Manias Panics and Crashes A History of Financial Crises 7th download pdf

Download the full version of the ebook at
https://ebookultra.com
An Introduction to Applied Multivariate

Analysis 1st Edition Tenko Raykov
https://ebookultra.com/download/an-introduction-
to-applied-multivariate-analysis-1st-edition-
tenko-raykov/
Explore and download more ebook at https://ebookultra.com

Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.
An Introduction to Multivariate Statistical Analysis Wiley

Series in Probability and Statistics 3rd Edition T. W.
Anderson
https://ebookultra.com/download/an-introduction-to-multivariate-
statistical-analysis-wiley-series-in-probability-and-statistics-3rd-
edition-t-w-anderson/
ebookultra.com
Introduction to Applied Statistical Signal Analysis 3rd

Edition Shiavi
https://ebookultra.com/download/introduction-to-applied-statistical-
signal-analysis-3rd-edition-shiavi/
ebookultra.com
An introduction to applied linguistics Third Edition.

Edition Michael Ph Rodgers
https://ebookultra.com/download/an-introduction-to-applied-
linguistics-third-edition-edition-michael-ph-rodgers/
ebookultra.com
An Introduction to Harmonic Analysis 3rd Edition Yitzhak

Katznelson
https://ebookultra.com/download/an-introduction-to-harmonic-
analysis-3rd-edition-yitzhak-katznelson/
ebookultra.com
Solutions Manual for An Introduction To Fourier Analysis
1st Edition Russell
https://ebookultra.com/download/solutions-manual-for-an-introduction-
to-fourier-analysis-1st-edition-russell/
ebookultra.com
An Introduction to Statistical Genetic Data Analysis

Melinda C. Mills
https://ebookultra.com/download/an-introduction-to-statistical-
genetic-data-analysis-melinda-c-mills/
ebookultra.com
Analysis of Multivariate Social Science Data Second

Edition Bartholomew
https://ebookultra.com/download/analysis-of-multivariate-social-
science-data-second-edition-bartholomew/
ebookultra.com
Introduction to Stochastic Calculus Applied to Finance

Second Edition Lamberton
https://ebookultra.com/download/introduction-to-stochastic-calculus-
applied-to-finance-second-edition-lamberton/
ebookultra.com
Cluster Analysis and Data Mining An Introduction King
https://ebookultra.com/download/cluster-analysis-and-data-mining-an-
introduction-king/
ebookultra.com
An Introduction to Applied Multivariate Analysis 1st
Edition Tenko Raykov Digital Instant Download
Author(s): Tenko Raykov, George A. Marcoulides
ISBN(s): 9781410618368, 1410618366
Edition: 1
File Details: PDF, 2.08 MB
Year: 2008
Language: english
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page i 2.2.2008 2:54pm Compositor Name: BMani
An Introduction
to Applied
Multivariate
Analysis
Tenko Raykov George A. Marcoulides
New York London

Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page ii 2.2.2008 2:54pm Compositor Name: BMani
Routledge Routledge
Taylor & Francis Group Taylor & Francis Group
270 Madison Avenue 2 Park Square
New York, NY 10016 Milton Park, Abingdon
Oxon OX14 4RN
© 2008 by Taylor & Francis Group, LLC

Routledge is an imprint of Taylor & Francis Group, an Informa business
Printed in the United States of America on acid-free paper

10 9 8 7 6 5 4 3 2 1
International Standard Book Number-13: 978-0-8058-6375-8 (Hardcover)
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, trans-
mitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter
invented, including photocopying, microfilming, and recording, or in any information storage or retrieval
system, without written permission from the publishers.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Introduction to applied multivariate analysis / by Tenko Raykov & George A.

Marcoulides.
p. cm.
Includes bibliographical references and index.
ISBN-13: 978-0-8058-6375-8 (hardcover)
ISBN-10: 0-8058-6375-3 (hardcover)
1. Multivariate analysis. I. Raykov, Tenko. II. Marcoulides, George A.
QA278.I597 2008
519.5’35--dc22 2007039834
Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com
and the Psychology Press Web site at

http://www.psypress.com
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page iii 2.2.2008 2:54pm Compositor Name: BMani
Contents
Preface ............................................................................................................... ix
Chapter 1 Introduction to Multivariate Statistics

1.1 Definition of Multivariate Statistics ............................................. 1
1.2 Relationship of Multivariate Statistics
to Univariate Statistics .................................................................. 5
1.3 Choice of Variables and Multivariate Method,
and the Concept of Optimal Linear Combination ....................... 7
1.4 Data for Multivariate Analyses .................................................... 8
1.5 Three Fundamental Matrices in Multivariate Statistics ............ 11
1.5.1 Covariance Matrix...................................................................... 12
1.5.2 Correlation Matrix...................................................................... 13
1.5.3 Sums-of-Squares and Cross-Products Matrix ........................ 15
1.6 Illustration Using Statistical Software........................................ 17
Chapter 2 Elements of Matrix Theory

2.1 Matrix Definition ......................................................................... 31
2.2 Matrix Operations, Determinant, and Trace .............................. 33
2.3 Using SPSS and SAS for Matrix Operations .............................. 46
2.4 General Form of Matrix Multiplications With Vector,
and Representation of the Covariance, Correlation,
and Sum-of-Squares and Cross-Product Matrices ..................... 50
2.4.1 Linear Modeling and Matrix Multiplication .......................... 50
2.4.2 Three Fundamental Matrices of Multivariate Statistics
in Compact Form ....................................................................... 51
2.5 Raw Data Points in Higher Dimensions, and Distance
Between Them.............................................................................. 54
Chapter 3 Data Screening and Preliminary Analyses

3.1 Initial Data Exploration............................................................... 61
3.2 Outliers and the Search for Them............................................... 69
3.2.1 Univariate Outliers..................................................................... 69
3.2.2 Multivariate Outliers ................................................................. 71
3.2.3 Handling Outliers: A Revisit .................................................... 78
3.3 Checking of Variable Distribution Assumptions....................... 80
3.4 Variable Transformations............................................................ 83
iii
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page iv 2.2.2008 2:54pm Compositor Name: BMani
iv
Chapter 4 Multivariate Analysis of Group Differences

4.1 A Start-Up Example .................................................................... 99
4.2 A Definition of the Multivariate Normal Distribution ............ 101
4.3 Testing Hypotheses About a Multivariate Mean..................... 102
4.3.1 The Case of Known Covariance Matrix................................ 103
4.3.2 The Case of Unknown Covariance Matrix........................... 107
4.4 Testing Hypotheses About Multivariate Means of
Two Groups ............................................................................... 110
4.4.1 Two Related or Matched Samples
(Change Over Time) ................................................................ 110
4.4.2 Two Unrelated (Independent) Samples ................................ 113
4.5 Testing Hypotheses About Multivariate Means
in One-Way and Higher Order Designs (Multivariate
Analysis of Variance, MANOVA)............................................. 116
4.5.1 Statistical Significance Versus Practical Importance ........... 129
4.5.2 Higher Order MANOVA Designs ......................................... 130
4.5.3 Other Test Criteria ................................................................... 132
4.6 MANOVA Follow-Up Analyses ............................................... 143
4.7 Limitations and Assumptions of MANOVA............................ 145
Chapter 5 Repeated Measure Analysis of Variance

5.1 Between-Subject and Within-Subject Factors
and Designs................................................................................ 148
5.2 Univariate Approach to Repeated Measure Analysis ............. 150
5.3 Multivariate Approach to Repeated Measure Analysis .......... 168
5.4 Comparison of Univariate and Multivariate Approaches
to Repeated Measure Analysis.................................................. 179
Chapter 6 Analysis of Covariance

6.1 Logic of Analysis of Covariance ............................................... 182
6.2 Multivariate Analysis of Covariance........................................ 192
6.3 Step-Down Analysis (Roy–Bargmann Analysis) ..................... 198
6.4 Assumptions of Analysis of Covariance .................................. 203
Chapter 7 Principal Component Analysis

7.1 Introduction ............................................................................... 211
7.2 Beginnings of Principal Component Analysis ......................... 213
7.3 How Does Principal Component Analysis Proceed?............... 220
7.4 Illustrations of Principal Component Analysis ....................... 224
7.4.1 Analysis of the Covariance Matrix S (S) of the
Original Variables .................................................................... 224
7.4.2 Analysis of the Correlation Matrix P (R) of the
Original Variables .................................................................... 224
7.5 Using Principal Component Analysis in Empirical Research......... 234
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page v 2.2.2008 2:54pm Compositor Name: BMani
7.5.1 Multicollinearity Detection ..................................................... 234

7.5.2 PCA With Nearly Uncorrelated Variables Is
Meaningless............................................................................... 235
7.5.3 Can PCA Be Used as a Method for Observed Variable
Elimination? .............................................................................. 236
7.5.4 Which Matrix Should Be Analyzed? ..................................... 236
7.5.5 PCA as a Helpful Aid in Assessing Multinormality .......... 237
7.5.6 PCA as ‘‘Orthogonal’’ Regression ......................................... 237
7.5.7 PCA Is Conducted via Factor Analysis Routines in
Some Software .......................................................................... 237
7.5.8 PCA as a Rotation of Original Coordinate Axes................. 238
7.5.9 PCA as a Data Exploratory Technique ................................. 238
Chapter 8 Exploratory Factor Analysis

8.1 Introduction ............................................................................... 241
8.2 Model of Factor Analysis .......................................................... 242
8.3 How Does Factor Analysis Proceed?........................................ 248
8.3.1 Factor Extraction ...................................................................... 248
8.3.1.1 Principal Component Method................................. 248
8.3.1.2 Maximum Likelihood Factor Analysis................... 256
8.3.2 Factor Rotation ......................................................................... 262
8.3.2.1 Orthogonal Rotation ................................................. 266
8.3.2.2 Oblique Rotation ....................................................... 267
8.4 Heywood Cases ......................................................................... 273
8.5 Factor Score Estimation............................................................. 273
8.5.1 Weighted Least Squares Method
(Generalized Least Squares Method) .................................... 274
8.5.2 Regression Method .................................................................. 274
8.6 Comparison of Factor Analysis and Principal
Component Analysis ................................................................. 276
Chapter 9 Confirmatory Factor Analysis

9.1 Introduction ............................................................................... 279
9.2 A Start-Up Example .................................................................. 279
9.3 Confirmatory Factor Analysis Model ....................................... 281
9.4 Fitting Confirmatory Factor Analysis Models ......................... 284
9.5 A Brief Introduction to Mplus, and Fitting the
Example Model ......................................................................................... 287
9.6 Testing Parameter Restrictions in Confirmatory Factor
Analysis Models ........................................................................ 298
9.7 Specification Search and Model Fit Improvement................... 300
9.8 Fitting Confirmatory Factor Analysis Models to the
Mean and Covariance Structure ............................................... 307
9.9 Examining Group Differences on Latent Variables ................. 314
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page vi 2.2.2008 2:54pm Compositor Name: BMani
vi
Chapter 10 Discriminant Function Analysis

10.1 Introduction ............................................................................. 331
10.2 What Is Discriminant Function Analysis?.............................. 332
10.3 Relationship of Discriminant Function Analysis to Other
Multivariate Statistical Methods............................................. 334
10.4 Discriminant Function Analysis With Two Groups .............. 336
10.5 Relationship Between Discriminant Function and
Regression Analysis With Two Groups ................................. 351
10.6 Discriminant Function Analysis With More Than
Two Groups ............................................................................. 353
10.7 Tests in Discriminant Function Analysis ............................... 355
10.8 Limitations of Discriminant Function Analysis ..................... 364
Chapter 11 Canonical Correlation Analysis

11.1 Introduction ............................................................................. 367
11.2 How Does Canonical Correlation Analysis Proceed? ........... 370
11.3 Tests and Interpretation of Canonical Variates ..................... 372
11.4 Canonical Correlation Approach to Discriminant
Analysis .................................................................................... 384
11.5 Generality of Canonical Correlation Analysis ....................... 389
Chapter 12 An Introduction to the Analysis

of Missing Data
12.1 Goals of Missing Data Analysis .............................................. 391
12.2 Patterns of Missing Data ......................................................... 392
12.3 Mechanisms of Missing Data .................................................. 394
12.3.1 Missing Completely at Random ........................................ 396
12.3.2 Missing at Random .............................................................. 398
12.3.3 Ignorable Missingness and Nonignorable
Missingness Mechanisms .................................................... 400
12.4 Traditional Ways of Dealing With Missing Data ...................401
12.4.1 Listwise Deletion .................................................................. 402
12.4.2 Pairwise Deletion ................................................................. 402
12.4.3 Dummy Variable Adjustment ............................................ 403
12.4.4 Simple Imputation Methods............................................... 403
12.4.5 Weighting Methods ............................................................. 405
12.5 Full Information Maximum Likelihood
and Multiple Imputation ......................................................... 406
12.6 Examining Group Differences and Similarities
in the Presence of Missing Data...............................................407
12.6.1 Examining Group Mean Differences With
Incomplete Data ................................................................... 410
12.6.2 Testing for Group Differences in the Covariance
and Correlation Matrices With Missing Data .................. 427
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page vii 2.2.2008 2:54pm Compositor Name: BMani
vii
Chapter 13 Multivariate Analysis of Change Processes

13.1 Introduction ............................................................................. 433
13.2 Modeling Change Over Time With Time-Invariant
and Time-Varying Covariates ................................................. 434
13.2.1 Intercept-and-Slope Model ................................................. 435
13.2.2 Inclusion of Time-Varying and Time-Invariant
Covariates.............................................................................. 436
13.2.3 An Example Application..................................................... 437
13.2.4 Testing Parameter Restrictions........................................... 442
13.3 Modeling General Forms of Change Over Time.....................448
13.3.1 Level-and-Shape Model....................................................... 448
13.3.2 Empirical Illustration ........................................................... 450
13.3.3 Testing Special Patterns of Growth or Decline................ 455
13.3.4 Possible Causes of Inadmissible Solutions ....................... 459
13.4 Modeling Change Over Time With Incomplete Data ............ 461
Appendix: Variable Naming and Order for Data Files ............ 467
References.......................................................................................... 469
Author Index ..................................................................................... 473
Subject Index ..................................................................................... 477

Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page viii 2.2.2008 2:54pm Compositor Name: BMani
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page ix 2.2.2008 2:54pm Compositor Name: BMani
Preface
Having taught applied multivariate statistics for a number of years, we

have been impressed by the broad spectrum of topics that one may be
expected to typically cover in a graduate course for students from depart-
ments outside of mathematics and statistics. Multivariate statistics has
developed over the past few decades into a very extensive field that is
hard to master in a single course, even for students aiming at methodo-
logical specialization in commonly considered applied fields, such as
those within the behavioral, social, and educational disciplines. To meet
this challenge, we tried to identify a core set of topics in multivariate
statistics, which would be both of fundamental relevance for its under-
standing and at the same time would allow the student to move on to
more advanced pursuits.
This book is a result of this effort. Our goal is to provide a coherent
introduction to applied multivariate analysis, which would lay down the
basics of the subject that we consider of particular importance in many
empirical settings in the social and behavioral sciences. Our approach is
based in part on emphasizing, where appropriate, analogies between
univariate statistics and multivariate statistics. Although aiming, in prin-
ciple, at a relatively nontechnical introduction to the subject, we were not
able to avoid the use of mathematical formulas, but we employ these
primarily in their definitional meaning rather than as elements of proofs
or related derivations. The targeted audience who will find this book most
beneficial consists primarily of graduate students, advanced undergradu-
ate students, and researchers in the behavioral, social, as well as educa-
tional disciplines, who have limited or no familiarity with multivariate
statistics. As prerequisites for this book, an introductory statistics course
with exposure to regression analysis is recommended, as is some famil-
iarity with two of the most widely circulated statistical analysis software:
SPSS and SAS.
Without the use of computers, we find that an introduction to applied
multivariate statistics is not possible in our technological era, and so we
employ extensively these popular packages, SPSS and SAS. In addition,
for the purposes of some chapters, we utilize the latent variable modeling
program Mplus, which is increasingly used across the social and behav-
ioral sciences. On the book specific website, www.psypress.com=applied-
multivariate-analysis, we supply essentially all data used in the text. (See
Appendix for name of data file and of its variables, as well as their order as
ix
Raykov/Introduction to Applied Multivariate Analysis RT20712_C000 Final Proof page x 2.2.2008 2:54pm Compositor Name: BMani
columns within it.) To aid with clarity, the software code (for SAS and
Mplus) or sequence of analytic=menu option selection (for SPSS) is also
presented and discussed at appropriate places in the book.
We hope that readers will find this text offering them a useful introduc-
tion to and a basic treatment of applied multivariate statistics, as well as
preparing them for more advanced studies of this exciting and compre-
hensive subject. A feature that seems to set apart the book from others in
this field is our use of latent variable modeling in later chapters to address
some multivariate analysis questions of special interest in the behavioral
and social disciplines. These include the study of group mean differences
on unobserved (latent) variables, testing of latent structure, and some
introductory aspects of missing data analysis and longitudinal modeling.
Many colleagues have at least indirectly helped us in our work on this
project. Tenko Raykov acknowledges the skillful introduction to multi-
variate statistics years ago by K. Fischer and R. Griffiths, as well as many
valuable discussions on the subject with S. Penev and Y. Zuo. George A.
Marcoulides is most grateful to H. Loether, B. O. Muthén, and D. Nasatir
under whose stimulating tutelage many years ago he was first introduced
to multivariate analysis. We are also grateful to C. Ames and R. Prawat
from Michigan State University for their instrumental support in more
than one way, which allowed us to embark on the project of writing this
book. Thanks are also due to L. K. Muthén, B. O. Muthén, T. Asparouhov,
and T. Nguyen for valuable instruction and discussions on applications
of latent variable modeling. We are similarly grateful to P. B. Baltes,
F. Dittmann-Kohli, and R. Kliegl for generously granting us access to
data from their project ‘‘Aging and Plasticity in Fluid Intelligence,’’ parts
of which we adopt for our method illustration purposes in several
chapters of the book. Many of our students provided us with very useful
feedback on the lecture notes we first developed for our courses in applied
multivariate statistics, from which this book emerged. We are also very
grateful to Douglas Steinley, University of Missouri-Columbia; Spiridon
Penev, University of New South Wales; and Tim Konold, University of
Virginia for their critical comments on an earlier draft of the manuscript,
as well as to D. Riegert and R. Larsen from Lawrence Erlbaum Associates,
and R. Tressider of Taylor & Francis, for their essential assistance during
advanced stages of our work on this project. Last but not least, we are
more than indebted to our families for their continued support in lots of
ways. Tenko Raykov thanks Albena and Anna, and George A. Marcoulides
thanks Laura and Katerina.
Tenko Raykov
East Lansing, Michigan
George A. Marcoulides
Riverside, California
Raykov/Introduction to Applied Multivariate Analysis RT20712_C001 Final Proof page 1 30.1.2008 4:37pm Compositor Name: BMani
1
Introduction to Multivariate Statistics
One of the simplest conceptual definitions of multivariate statistics (MVS)

is as a set of methods that deal with the simultaneous analysis of multiple
outcome or response variables, frequently also referred to as dependent
variables (DVs). This definition of MVS suggests an important relationship
to univariate statistics (UVS) that may be considered a group of methods
dealing with the analysis of a single DV. In fact, MVS not only exhibits
similarities with UVS but can also be considered an extension of it, or
conversely UVS can be viewed as a special case of MVS. At the same time,
MVS and UVS have a number of distinctions, and this book deals with
many of them whenever appropriate.
In this introductory chapter, our main objective is to discuss, from a
principled standpoint, some of the similarities and differences between
MVS and UVS. More specifically, we (a) define MVS; then (b) discuss
some relationships between MVS and UVS; and finally (c) illustrate the
use of the popular statistical software SPSS and SAS for a number of initial
multiple variable analyses, including obtaining covariance, correlation,
and sum-of-squares and cross-product matrices. As will be observed
repeatedly throughout the book, these are three matrices of variable
interrelationship indices, which play a fundamental role in many MVS
methods.
1.1 Definition of Multivariate Statistics

Behavioral, social, and educational phenomena are often multifaceted,
multifactorially determined, and exceedingly complex. Any systematic
attempt to understand them, therefore, will typically require the examin-
ation of multiple dimensions that are usually intertwined in complicated
ways. For these reasons, researchers need to evaluate a number of inter-
related variables capturing specific aspects of phenomena under consid-
eration. As a result, scholars in these sciences commonly obtain and have
to deal with data sets that contain measurements on many interdependent
dimensions, which are collected on subjects sampled from the studied
1
2 Introduction to Applied Multivariate Analysis
populations. Consequently, in empirical behavioral and social studies, one

is very often faced with data sets consisting of multiple interrelated
variables that have been observed on a number of persons, and possibly
on samples from different populations.
MVS is a scientific field, which for many purposes may be viewed a
branch of mathematics and has been developed to meet these complex
challenges. Specifically, MVS represents a multitude of statistical methods
to help one analyze potentially numerous interrelated measures consid-
ered together rather than separately from one another (i.e., one at a time).
Researchers typically resort to using MVS methods when they need to
analyze more than one dependent (response, or outcome) variable, pos-
sibly along with one or more independent (predictor, or explanatory)
variables, which are in general all correlated with each other.
Although the concepts of independent variables (IVs) and DVs are
generally well covered in most introductory statistics and research
methods treatments, for the aims of this chapter, we deem it useful to
briefly discuss them here. IVs are typically different conditions to which
subjects might be exposed, or reflect specific characteristics that studied
persons bring into a research situation. For example, socioeconomic sta-
tus (SES), educational level, age, gender, teaching method, training pro-
gram or treatment are oftentimes considered IVs in various empirical
settings. Conversely, DVs are those that are of main interest in an inves-
tigation, and whose examination is of focal interest to the researcher. For
example, intelligence, aggression, college grade point average (GPA) or
Graduate Record Exam (GRE) score, performance on a reading or writing
test, math ability score or computer aptitude score can be DVs in a study
aimed at explaining variability in any of these measures in terms of some
selected IVs. More specifically, the IVs and DVs are defined according to
the research question being asked. For this reason, it is possible that a
variable that is an IV for one research query may become a DV for
another one, or vice versa. Even within a single study, it is not unlikely
that a DV for one question of interest changes status to an IV, or
conversely, when pursuing another concern at a different point during
the study.
To give an example involving IVs and DVs: suppose an educational
scientist were interested in comparing the effectiveness of two teaching
methods, a standard method and a new method of teaching number
division. To this end, two groups of students are randomly assigned to
the new and to the standard method. Assume that a test of number
division ability was administered to all students who participated in the
study, and that the researcher was interested in explaining the individual
differences observed then. In this case, the score on the division test would
be a DV. If the scientist had measured initial arithmetic ability as well as
collected data on student SES or even hours watching television per week
then all these three variables may be potential IVs. The particular posited
Introduction to Multivariate Statistics 3
question appears to be relatively simple and in fact may be addressed

straightforwardly using UVS as it is phrased in terms of a single DV,
namely score obtained on the number division test. However, if the
study was carried out in such a way that measurements of division ability
were collected for each student on each of three consecutive weeks after
the two teaching methods were administered, and in addition data on
hours watched television were gathered in each of these weeks, then this
question becomes considerably more complicated. This is because there
are now three measures of interest—division ability in each of the 3 weeks
of measurement—and it may be appropriate to consider them all as DVs.
Furthermore, because these measures are taken on the same subjects in the
study, they are typically interrelated. In addition, when addressing the
original question about comparative effectiveness of the new teaching
method relative to the old one, it would make sense at least in some
analyses to consider all three so-obtained division ability scores simultan-
eously. Under these circumstances, UVS cannot provide the sought
answer to the research query. This is when MVS is typically used, espe-
cially where the goal is to address complicated research questions that
cannot be answered directly with UVS.
A main reason for this MVS preference in such situations is that UVS
represents a set of statistical methods to deal with just a single DV, while
there is effectively no limit on the number of IVs that might be considered.
As an example, consider the question of whether observed individual
differences in average university freshman grades could be explained
with such on their SAT score. In this case, the DV is freshman GPA,
while the SAT score would play the role of an IV, possibly in addition to
say gender, SES and type of high school attended (e.g., public vs. private),
in case a pursued research question requires consideration of these as
further IVs.
There are many UVS and closely related methods that can be used to
address an array of queries varying in their similarity to this one. For
example, for prediction goals, one could consider using regression analy-
sis (simple or multiple regression, depending on the number of IVs
selected), including the familiar t test. When examination of mean differ-
ences across groups is of interest, a traditionally utilized method would be
analysis of variance (ANOVA) as well as analysis of covariance
(ANCOVA)—either approach being a special case of regression analysis.
Depending on the nature of available data, one may consider a chi-square
analysis of say two-way frequency tables (e.g., for testing association of
two categorical variables). When certain assumptions are markedly vio-
lated, in particular the assumption of normality, and depending on other
study aspects, one may also consider nonparametric statistical methods.
Most of the latter methods share the common feature that they are typic-
ally considered for application when for certain research questions one
identifies single DVs.
By way of contrast, MVS may be viewed as an extension of UVS in the

case where one is interested in studying multiple DVs that are interrelated,
as they commonly would be in empirical research in the behavioral, social,
and educational disciplines. For this reason, MVS typically deals in appli-
cations with fairly large data sets on potentially many subjects and in
particular on possibly numerous interrelated variables of main interest.
Due to this complexity, the underlying theme behind many MVS methods
is also simplification, for instance, the reduction of the complexity of
available data to several meaningful indices, quantities (parameters), or
dimensions.
To give merely a sense of the range of questions addressed with MVS,
let us consider a few simple examples—we of course return to these issues
in greater detail later in this book. Suppose a researcher is interested in
determining which characteristics or variables differentiate between
achievers and nonachievers in an educational setting. As will be discussed
in Chapter 10, these kinds of questions can be answered using a technique
called discriminant function analysis (or discriminant analysis for short).
Interestingly, discriminant analysis can also be used to predict group
membership—in the currently considered example, achievers versus
nonachievers—based on the knowledge obtained about differentiating
characteristics or variables. Another research question may be whether
there is a single underlying (i.e., unobservable, or so-called latent) dimen-
sion along which students differ and which is responsible for their
observed interrelationships on some battery of intelligence tests. Such a
question can be attended to using a method called factor analysis (FA),
discussed in Chapters 8 and 9. As another example, one may be concerned
with finding out whether a set of interrelated tests can be decomposed into
groups of measures and accordingly new derived measures obtained so
that they account for most of the observed variance of the tests. For these
aims, a method called principal component analysis (PCA) is appropriate,
to which we turn in Chapter 7. Further, if one were interested in whether
there are mean differences between several groups of students exposed to
different training programs, say with regard to their scores on a set of
mathematical tasks—possibly after accounting for initial differences on
algebra, geometry, and trigonometry tests—then multivariate analysis of
variance (MANOVA) and multivariate analysis of covariance (MAN-
COVA) would be applicable. These methods are the subject of Chapters
4 and 6. When of concern are group differences on means of unobserved
variables, such as ability, intelligence, neuroticism, or aptitude, a specific
form of what is referred to as latent variable modeling could be used
(Chapter 9). Last but not least, when studied variables have been repeat-
edly measured, application of special approaches of ANOVA or latent
variable modeling can be considered, as covered in Chapters 5 and 13. All
these examples are just a few of the kinds of questions that can be
addressed using MVS, and the remaining chapters in this book are
devoted to their discussion. The common theme unifying the research

questions underlying these examples is the necessity to deal with poten-
tially multiple correlated variables in such a way that their interrelation-
ships are taken into account rather than ignored.
1.2 Relationship of Multivariate Statistics

to Univariate Statistics
The preceding discussion provides leads to elaborate further on the rela-
tionship between MVS and UVS. First, as indicated previously, MVS may
be considered an extension of UVS to the case of multiple, and commonly
interrelated, DVs. Conversely, UVS can be viewed as a special case of
MVS, which is obtained when the number of analyzed DVs is reduced to
just one. This relationship is additionally highlighted by the observation
that for some UVS methods, there is an MVS analog or multivariate gener-
alization. For example, traditional ANOVA is extended to MANOVA
in situations involving more than one outcome variable. Similarly, con-
ventional ANCOVA is generalized to MANCOVA whenever more than a
single DV is examined, regardless of number of covariates involved.
Further, multiple regression generalizes to multivariate multiple regres-
sion (general linear model) in the case with more than one DVs. This type
of regression analysis may also be viewed as path analysis or structural
equation modeling with observed variables only, for which we refer to a
number of alternative treatments in the literature (see Raykov & Marcou-
lides, 2006, for an introduction to the subject). Also, the idea underlying
the widely used correlation coefficient, for example, in the context of a
bivariate correlation analysis or simple linear regression, is extended to
that of canonical correlation. In particular, using canonical correlation
analysis (CCA) one may examine the relationships between sets of what
may be viewed, for the sake of this example, as multiple IVs and multiple
DVs. With this perspective, CCA could in fact be considered encompass-
ing all MVS methods mentioned so far in this section, with the latter being
obtained as specifically defined special cases of CCA.
With multiple DVs, a major distinctive characteristic of MVS relative to
UVS is that the former lets one perform a single, simultaneous analysis
pertaining to the core of a research question. This approach is in contrast
to a series of univariate or even bivariate analyses, like regressions with a
single DV, correlation estimation for all pairs of analyzed variables, or
ANOVA=ANCOVA for each DV considered in turn (i.e., one at a time).
Even though we often follow such a simultaneous multivariate test with
further and more focused analyses, the benefit of using MVS is that no
matter how many outcome variables are analyzed the overall Type I error
rate is kept at a prespecified level, usually .05 or more generally the one at
which the multivariate test is carried out. As a conceivable alternative, one
might contemplate conducting multiple univariate analyses, one per DV.
However, that approach will be associated with a higher (family-wise)
Type I error rate due to the multiple testing involved. These are essentially
the same reasons for which in a group mean comparison setup, carrying
out a series of t tests for all pairs of groups would be associated with a
higher than nominal error rate relative to an ANOVA, and hence make the
latter a preferable analytic procedure.
At the same time, it is worth noting that with MVS we aim at the ‘‘big
picture,’’ namely analysis of more than one DV when considered together.
This is why with MVS we rarely get ‘‘as close’’ to data as we can with
UVS, because we typically do not pursue as focused an analysis in MVS
as we do in UVS where a single DV is of concern. We emphasize however
that the center of interest, and thus of analysis, depends on the specific
research question asked. For example, at any given stage of an empirical
study dealing with say a teaching method comparison, we may be inter-
ested in comparing two or more methods with regard only to a
single DV. In such a case, the use of UVS will be quite appropriate.
When alternatively the comparison is to be carried out with respect to
several DVs simultaneously, an application of MVS is clearly indicated
and preferable.
In conclusion of this section, and by way of summarizing much of the
preceding discussion, multivariate analyses are conducted instead of uni-
variate analyses for the following reasons:
1. With more than one DVs (say p in number), the use of p separate
univariate tests inflates the Type I error rate, whereas a pertinent
multivariate test preserves the significance level (p > 1).
2. Univariate tests, no matter how many in number, ignore the
interrelationships possible among the DVs, unlike multivariate
analyses, and hence potentially waste important information con-
tained in the available sample of data.
3. In many cases, the multivariate test is more powerful than a
corresponding univariate test, because the former utilizes the
information mentioned in the previous point 2. In such cases, we
tend to trust MVS more when its results are at variance with those
of UVS (as we also do when of course our concern is primarily
with a simultaneous analysis of more than one DV).
4. Many multivariate tests involving means have as a by-product the
construction of a linear combination of variables, which provides
further information (in case of a significant outcome) about how
the variables unite to reject the hypothesis; we deal with these
issues in detail later in the book (Chapter 10).
1.3 Choice of Variables and Multivariate Method,

and the Concept of Optimal Linear Combination
Our discussion so far has assumed that we have already selected the
variables to be used in a given multivariate analysis. The natural question
that arises now is how one actually makes this variable choice. In general,
main requirements for the selection of variables to be used in an MVS
analysis—like in any univariate analysis—are those of high psychometric
qualities (specifically, high validity and reliability) of the measures used as
DVs and IVs, and that they pertain to the research questions being pursued,
that is, are measures of aspects of a studied phenomenon that are relevant to
these questions. Accordingly, throughout the rest of the book, we assume
that the choice of considered variables has already been made in this way.
MVS encompasses an impressive array of analytic and modeling
methods, each of which can be used to tackle certain research queries.
Consequently, the next logical concern for any researcher is which one(s)
of these methods to utilize. In addition to selecting from among those
methods that allow answering the questions asked, the choice will also
typically depend on the type of measurement of the involved DVs. Often-
times, with continuous (or approximately so) DVs, such as reaction time,
income, GPA, and intelligence test scores, a frequent choice may be made
from among MANOVA or MANCOVA, FA, PCA, discriminant function
analysis, CCA, or multivariate multiple regression. Alternatively, with
discrete DVs—for example, answers to questions from a questionnaire
that have limited number of response options, or items of another type
of multiple-component measuring instrument—a choice may often be
made from among logistic regression, contingency table analysis, log-
linear models, latent variable modeling with categorical outcomes (e.g.,
latent class analysis), or item–response theory models. Obviously, even
within a comprehensive textbook, only a limited number of these methods
can be adequately addressed. Having to make such a choice, we elected
the material in the book to center around what we found to be—at the
level aimed in this text—most widely used multivariate methods for
analysis of continuous DVs. For discussions of methods for analyzing
discrete DVs, the reader is referred to a number of alternative texts
(Agresti, 2002; Lord, 1980; Muthén, 2002; Skrondal & Rabe-Hesketh, 2004).
When choosing a statistical technique, in particular a multivariate
method, whether the data to be analyzed are experimental (i.e., resulting
after random assignment of subjects and manipulation of IVs, in addition to
typically exercising control upon so-called extraneous variance) or obser-
vational (i.e., obtained from responses to questionnaires or surveys), is
irrelevant. Statistical analysis will work in either case equally well.
However, it is the resultant interpretation which will typically differ. In
particular, potential causality attributions, if attempted, can be crucially
affected by whether the data stem from an experimental or an observational

(correlational) study. A researcher is in the strongest position to possibly
make causal statements in the case of an experimental study. This funda-
mental matter is discussed at length in the literature, and we refer the reader
to alternative sources dealing with experimental design, causality and
related issues (Shadish, Cook, & Campbell, 2002). Throughout this book,
we will not make or imply causal statements of any form or type.
Once the particular choice of variables and statistical method(s) of analy-
sis is made, a number of MVS techniques will optimally combine the DVs
and yield a special linear combination of them with certain features of
interest. More specifically, these methods find that linear combination, Y*,
of the response measures Y1, Y2, . . . , Yp (p > 1), which is defined as
Y* ¼ w1 Y1 þ w2 Y2 þ þ wp Yp (1:1)
and has special optimality properties (Y* is occasionally referred to as

supervariable). In particular, this constructed variable Y* may be best at
differentiating between groups of subjects that are built with respect to
some IV(s). As discussed in Chapter 10, this will be the case when using
the technique called discriminant function analysis. Alternatively, the
variable Y* defined in Equation 1.1 may possess the highest possible
variance from all linear combinations of the measures Y1, Y2, . . . , Yp that
one could come up with. As discussed in Chapter 7, this will be the case
when using PCA. As another option, Y* may be constructed so as to
possess the highest correlation with another appropriately obtained super-
variable, that is, a linear combination of another set of variables. This will
be the case in CCA. Such optimal linear combinations are typical for MVS,
and in some sense parallel the search conducted in univariate regression
analysis for that linear combination of a given set of predictors, which has
the highest possible correlation with a prespecified DV. Different MVS
methods use specific information about the relationship between the DVs
and possibly IVs in evaluating the weights w1, w2, . . . , wp in Equation 1.1,
so that the resulting linear combination Y* has the corresponding of the
properties mentioned above.
1.4 Data for Multivariate Analyses

In empirical behavioral, social, and educational research, MVS methods
are applied to data provided by examined subjects sampled from studied
populations. For the analytic procedures considered in this book, these
data are typically organized in what is commonly called data matrix.
Because the notion of a data matrix is of special relevance both in UVS
and MVS, we attend to it in some detail next.
TABLE 1.1
Data From Four Subjects in a General Mental Ability Study
Student Test1 Test2 Test3 Gen SES MathAbTest
1 45 55 47 1 3 33
2 51 54 57 0 1 23
3 40 51 46 1 2 43
4 49 45 48 0 3 42
Note: Gen ¼ gender; SES ¼ socioeconomic status; MathAbTest ¼ mathematics ability test
score.
A data matrix is a rectangular array of collected (recorded) scores from

studied subjects. The entries in this matrix are arranged in such a way that
each row represents a given person’s data, and each column represents a
variable that may be either dependent or independent in a particular
analysis, in accordance with a research question asked. For example,
consider a study of general mental ability, and let us assume that data
are collected from a sample of students on the following six variables: (a)
three tests of intelligence, denoted below Test1, Test2, and Test3; (b) a
mathematics ability test; (c) information about their gender; and (d) SES.
For the particular illustrative purposes here, the data on only four subjects
are provided in Table 1.1.
As seen from Table 1.1, for the continuous variables—in this case, the
three tests of intelligence and the mathematics ability test—the actual
performance scores are recorded in the data matrix. However, for discrete
variables (here, gender and SES), codes for group membership are typic-
ally entered. For more details on coding schemes, especially in studies
with multiple groups—a topic commonly discussed at length in most
regression analysis treatments—we refer to Pedhazur (1997) or any other
introductory to intermediate statistics text.
Throughout the rest of this book, we treat the elements of the data
matrix as discussed so far in this section. We note, however, that in some
repeated assessment studies (or mixed modeling contexts) data per meas-
urement occasion may be recorded on a single line. Such data arrange-
ments will not be considered in the book, yet we mention that they can be
readily reordered in the form of data matrices outlined above that will be
of relevance in the remainder. Additionally, this text will not be concerned
with data that, instead of being in the form of subject performance or
related scores, are determined as distances between stimuli or studied
persons (e.g., data collected in psychophysics experiments or some areas
of social psychology). For discussions on how to handle such settings, we
refer to corresponding sources (Johnson & Wichern, 2002).
The above data-related remarks also lead us to a major assumption that
will be made throughout the book, that of independence of studied
subjects. This assumption may be violated in cases where there is a

possibility for examined persons to interact with one another during the
process of data collection, or to receive the same type of schooling, treat-
ment instruction, or opportunities. The data resulting in such contexts
typically exhibit a hierarchical nature. For example, data stemming from
patients who are treated by the same doctor or in the same health care
facility, or data provided by students taught by the same teachers, in
certain classrooms or in schools (school districts), tend to possess this
characteristic. This data property is at times also referred to as nestedness
or clustering, and then scores from different subjects cannot be safely
considered independent. While the methods discussed in this book may
lead to meaningful results when the subject independence assumption is
violated to a minor extent, beyond that they cannot be generally trusted.
Instead, methods that have been specifically developed to deal with
nested data, also referred to as hierarchical, multilevel, or clustered data,
should be utilized. Such methods are available within the so-called mixed
modeling methodology, and some of them are also known as hierarchical
linear (or nonlinear) modeling. For a discussion of these methods, which
are not covered in this text, the reader is referred to Heck and Thomas
(2000), Hox (2002), or Raudenbush and Bryk (2002) and references therein.
Later chapters of this book will, however, handle a special case of hier-
archical data, which stem from repeated measure designs. In the latter,
measurements from any given subject can be considered nested within
that subject, that is, exhibit a two-level hierarchy. Beyond this relatively
simple case of nested structure, however, alternative multilevel analysis
approaches are recommended with hierarchical data.
While elaborating on the nature of the data utilized in applications of
MVS methods considered in this book, it is also worthwhile to make the
following notes. First, we stress that the rows of the data matrix (Table 1.1)
represent a random sample from a studied population, or random sam-
ples from more than one population in the more general case of a multi-
population investigation. That is, the rows of the data matrix are
independent of one another. On the other hand, the columns of the data
matrix do not represent a random sample, and are in general not inde-
pendent of one another but instead typically interrelated. MVS is in fact
particularly concerned with this information about the interrelationship
between variables (data matrix columns) of main interest in connection to
a research question(s) being pursued.
Second, in many studies in the behavioral, social, and educational dis-
ciplines, some subjects do not provide data on all collected variables.
Hence, a researcher has to deal with what is commonly referred to as
missing data. Chapter 12 addresses some aspects of this in general difficult
to deal with issue, but at this point we emphasize the importance of using
appropriate and uniform representation of missing values throughout the
entire data matrix. In particular, it is strongly recommended to use the
TABLE 1.2
Missing Data Declaration in an Empirical Study (cf. Table 1.1)
Student Test1 Test2 Test3 Gen SES MathAbTest
1 45 55 47 1 3 33
2 51 54 57 0 1 23
3 40 51 46 1 2 43
4 49 45 48 0 3 42
5 99 44 99 1 99 44
6 52 99 44 99 2 99
Note: Gen ¼ gender; SES ¼ socioeconomic status; MathAbTest ¼ mathematics ability test
score.
same symbol(s) for denoting missing data, a symbol(s) that is not a legit-
imate value possible to take by any subject on a variable in the study. In
addition, as a next step one should also insure that this value(s) is declared
to the software used as being employed to denote missing value(s); failure
to do so can cause severely misleading results. For example, if the next
two subjects in the previously considered general mental ability study
(cf. Table 1.1) had some missing data, the latter could be designated
by the uniform symbol (99) as illustrated in Table 1.2.
Dealing with missing data is in general a rather difficult and in part
‘‘technical’’ matter, and we refer the reader to Allison (2001) and Little and
Rubin (2002) for highly informative and instructive treatments of this issue
(see also Raykov, 2005, for a nontechnical introduction in a context of
repeated measure analysis). In this book, apart from the discussion in
Chapter 12, we assume that used data sets have no missing values (unless
otherwise indicated).
1.5 Three Fundamental Matrices in Multivariate Statistics

A number of MVS methods may be conceptually thought of as being
based on at least one of three matrices of variable interrelationship indices.
For a set of variables to be analyzed, denoted Y1, Y2, . . . , Yp (p > 1), these
matrices are
1. The covariance matrix, designated S for a given sample

2. The correlation matrix, symbolized R in the sample
3. The sum-of-squares and cross-products (SSCP) matrix, denoted Q
in the sample
For the purpose of setting the stage for subsequent developments, we

discuss each of these matrices in Sections 1.5.1 through 1.5.3.
1.5.1 Covariance Matrix

A covariance matrix is a symmetric matrix that contains the variable
variances on its main diagonal and the variable covariances as remaining
elements. The covariance coefficient represents a nonstandardized index
of the linear relationship between two variables (in case it is standardized,
this index is the correlation coefficient, see Section 1.5.2). For example,
consider an intelligence study of n ¼ 500 sixth-grade students, which used
five intelligence tests. Assume that the following was the resulting empir-
ical covariance matrix for these measures (for symmetry reasons, only
main diagonal elements and those below are displayed, a practice fol-
lowed throughout the remainder of the book):
2 3
75:73
6 23:55 66:77 7
6 7
6 7
S ¼ 6 33:11 37:22 99:54 7: (1:2)
6 7
4 29:56 33:41 37:41 74:44 5
21:99 31:25 22:58 33:66 85:32
We note that sample variances are always positive (unless of course on a

given variable all subjects take the same value, a highly uninteresting
case). In empirical research, covariances tend to be smaller, on average,
in magnitude (absolute value) than variances. In fact, this is a main
property of legitimate covariance matrices for observed variables of interest
in behavioral and social research. In Chapter 2, we return to this type of
matrices and discuss in a more formal way an important concept referred
to as positive definiteness. For now, we simply note that unless there is a
perfect linear relationship among a given set of observed variables (that
take on real values), their covariance matrix will exhibit this feature
informally mentioned here.
For a given data set on p variables (p > 1), there is a pertinent population
covariance matrix (on the assumption of existence of their variances,
which is practically not restrictive). This matrix, typically denoted S,
consists of the population variances on its main diagonal and population
covariances off it for all pairs of variables. In a random sample of n (n > 1)
subjects drawn from a studied population, each element of this covariance
matrix can be estimated as
Xn i )(Ykj Y
(Yki Y j)
sij ¼ (1:3)
k¼1
n1
(i, j ¼ 1, . . . , p), where the bar stands for arithmetic mean for the variable
underneath, while Yki and Ykj denote the score of the kth subject on the ith
and jth variable, respectively. (In this book, n generally denotes sample
size.) Obviously, in the special case that i ¼ j, from Equation 1.3 one
estimates the variable variances in that sample as
Xn i )2
(Yki Y
s2i ¼ , (1:4)
k¼1
n1
where (i ¼ 1, . . . , p). Equations 1.3 and 1.4 are utilized by statistical soft-
ware to estimate an empirical covariance matrix, once the data matrix with
observed raw data has been provided. If data were available on all
members of a given finite population (the latter being the typical case in
social and behavioral research), then using Equations 1.3 and 1.4 yet with
the divisor (denominator) n would allow one to determine all elements of
the population covariance matrix S.
We note that, as can be seen from Equations 1.3 and 1.4, variances and
covariances depend on the specific units of measurement of the variables
involved. In particular, the magnitude of either of these two indices is
unrestricted. Hence, a change from one unit of measurement to another
(such as from inches to centimeter measurements), which in empirical
research is usually done via a linear transformation, can substantially
increase or decrease the magnitude of variance and covariance coeffi-
cients.
1.5.2 Correlation Matrix

The last mentioned feature of scale dependence creates difficulties when
trying to interpret variances and covariances. To deal with this problem
for two random variables, one can use the correlation coefficient, which is
obtained by dividing their covariance with the square-rooted product of
their variances (i.e., the product of their standard deviations; see below for
the case when this division would not be possible). For a given population,
from the population covariance matrix S one can determine in this way
the population correlation matrix, which is commonly denoted R (the
capital Greek letter ‘‘rho’’). Similar to the covariance coefficient, in a
given sample the correlation coefficient rij between the variables Yi and
Yj is evaluated as
P
n
i )(Ykj Y j)
(Yki Y
sij sij k¼1
rij ¼ qffiffiffiffiffiffiffiffi ¼ ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (1:5)
si sj P i )2 P (Ykj Y
s2i s2j n n
(Yki Y j )2
k¼1 k¼1
(i, j ¼ 1, . . . , p). A very useful property of the correlation coefficient is that

the magnitude of the numerator in Equation 1.5 is never larger than that of
the denominator, that is,
sij si sj , (1:6)
and hence
1 rij 1 (1:7)
is always true (i, j ¼ 1, . . . , p). As a matter of fact, in Inequalities 1.6 and 1.7
the equality sign is only then obtained when there is a perfect linear
relationship between the two variables involved, that is, if and only
there exist two numbers aij and bij such that Yi ¼ aij þ bij Yj (1 < i, j < p).
As with the covariance matrix, if data were available on an entire (finite)
population, using Equation 1.5 one could determine the population cor-
relation matrix R. Also from Equation 1.5 it is noted that the correlation
coefficient will not exist, that is, will not be defined, if at least one of the
variables involved is constant across the sample or population considered
(this is obviously a highly uninteresting and potentially unrealistic case in
empirical research). Then obviously the commonly used notion of rela-
tionship is void as well.
As an example, consider a study examining the relationship between
GPA, SAT scores, annual family income, and abstract reasoning test scores
obtained from n ¼ 350, 10th-grade students. Let us also assume that in this
sample the following correlation matrix was obtained (positioning of
variables in the data matrix is as just listed, from left to right and top to
bottom):
2 3
1
6 :69 1 7
R¼6
4 :48
7:
5 (1:8)
:52 1
:75 :66 :32 1
As seen by examining the values in the right-hand side of Equation 1.8,

most correlations are of medium size, with that between income and
abstract reasoning test score being the weakest (.32). We note that while
there are no strict rules to be followed when interpreting entries in a
correlation matrix, it is generally easier to interpret their squared values
(e.g., as in simple linear regression, where the squared correlation equals
the R2 index furnishing the proportion of explained or shared variance
between the two variables in question).
We also observe that while in general only the sign of a covariance
coefficient can be interpreted, both the sign and magnitude of the correl-
ation coefficient are meaningful. Specifically, the closer the correlation
coefficient is to 1 or 1, the stronger (more discernible) the linear relation-

ship is between the two variables involved. In contrast, the closer this
coefficient is to 0, being either a positive or negative number, the weaker
the linear pattern of relationship between the variables. Like a positive
covariance, a positive correlation suggests that persons with above mean
(below mean) performance on one of the variables tend to be among those
with above mean (below mean) performance on the other measure. Alter-
natively, a negative covariance (negative correlation) is indicative of the
fact that subjects with above (below) mean performance on one of the
variables tend to be among those with below (above) mean performance
on the other measure. When the absolute value of the correlation coeffi-
cient is, for example, in the .90s, one may add that this tendency is strong,
while otherwise it is moderate to weak (the closer to 0 this correlation is).
1.5.3 Sums-of-Squares and Cross-Products Matrix

An important feature of both the covariance and correlation matrices is
that when determining or estimating their elements an averaging process
takes place (see the summation sign and division by (n 1) in Equations
1.3 through 1.5, and correspondingly for their population counterparts).
This does not happen, however, when one is concerned with estimating
the entries of another matrix of main relevance in MVS and especially
in ANOVA contexts, the so-called sums-of-squares and cross-products
(SSCP) matrix. For a sample from a studied population, this symmetric
matrix contains along its diagonal the sums of squares, and off this diagonal
the sums of cross products for all possible pairs of variables involved. As is
well known from discussions of ANOVA in introductory statistics text-
books, the sum of squares for a given variable is
X
n
qii ¼ i )2
(Yki Y (1:9)
k¼1
(i ¼ 1, . . . , p), which is sometimes also referred to as corrected or deviation

sum of squares, due to the fact that the mean is subtracted from the
observed score on the variable under consideration. Similarly, the sum
of cross products is
X
n
qij ¼ i )(Ykj Y
(Yki Y j ), (1:10)
k¼1
where 1 < i, j < p. Obviously, Equation 1.9 is obtained from Equation 1.10
in the special case when i ¼ j, that is, the sum of squares for a given
variable equals the sum of cross products of this variable with itself
(i, j ¼ 1, . . . , p). As can be readily seen by a comparison of Equations 1.9
and 1.10 with the Equations 1.3 and 1.4, respectively, the elements of the
SSCP matrix Q ¼ [qij] result by multiplying with (n 1) the corresponding
elements of the empirical covariance matrix S. (In the rest of this chapter,
we enclose in brackets the general element of a matrix; see Chapter 2 for
further notation explication.) Hence, Q has as its elements measures of
linear relationship that are not averaged over subjects, unlike the elements
of the matrices S and R. As a result, there is no readily conceptualized
population analogue of the SSCP matrix Q. This may in fact be one of the
reasons why this matrix has been referred to explicitly less often in the
literature (in particular applied) than the covariance or correlation matrix.
Another type of SSCP matrix that can also be considered and used to
obtain the matrix Q is one that reflects the SSCP of the actual raw scores
uncorrected for the mean. In this raw score SSCP matrix, U ¼ [uij], the sum
of squares for a given variable, say Yi is defined as
X
n
uii ¼ Y2ki , (1:11)
k¼1
and the sums of its cross products with the remaining variables are
X
n
uij ¼ Yki Ykj , (1:12)
k¼1
where 1 i, j p. We note that Equation 1.11 is obtained from Equation

1.12 in the special case when i ¼ j (i, j ¼ 1, . . . , p).
As an example, consider a study concerning the development of aggres-
sion in middle-school students, with p ¼ 3 consecutive assessments on an
aggression measure. Suppose the SSCP matrix for these three measures,
Q, is as follows:
2 3
1112:56
Q ¼ 4 992:76 2055:33 5: (1:13)
890:33 1001:36 2955:36
We observe that similarly to the covariance matrix, the diagonal entries of

Q are positive (unless a variable is a constant), and tend to be larger in
magnitude than the off-diagonal ones. This feature can be readily
explained with the fact mentioned earlier that the elements of Q equal
(n 1) times those of the covariance matrix S, and the similar feature of the
covariance matrix. We also note that the elements of Q may grow unre-
strictedly, or alternatively decrease unrestrictedly (if negative), when sam-
ple size increases; this is due to their definition as nonaveraged sums
across subjects. Similarly to the elements of any sample covariance matrix
S, due to their nonstandardized feature, the elements of Q cannot in
general be interpreted in magnitude, but only their sign can—in the same
way as the sign of the elements of S. Further, the entries in Q also depend
on the metric underlying the studied variables.
We stress that the matrices S, R, and Q will be very important in most
MVS methods of interest in this book because they contain information on
the linear interrelationships among studied variables. It is these interrela-
tionships, which are essential for the multivariate methods considered
later in the text. Specifically, MVS methods capitalize on this information
and re-express it in their results, in addition to other features of the data.
To illustrate, consider the following cases. A correlation matrix showing
uniformly high variable correlations (e.g., for a battery of tests) may reveal
a structure of relationships that is consistent with the assumption of a
single dimension (e.g., abstract thinking ability) underlying a set of ana-
lyzed variables. Further, a correlation matrix showing two groups of
similar (within-group) correlations with respect to size, may be consistent
with two interrelated dimensions (e.g., reading ability and writing ability
in an educational study). As it turns out, and preempting some of the
developments to follow in subsequent chapters, we use the correlation
and covariance matrix in FA and PCA; the SSCP matrix in MANOVA,
MANCOVA, and discriminant function analysis; and the covariance mat-
rix in confirmatory FA, in studies of group mean differences on unob-
served variables, and in such with repeated measures. (In addition to the
covariance matrix, also variable means will be of relevance in the latter
two cases, as elaborated in Chapters 9 and 13.)
Hence, with some simplification, we may say that these three matrices
of variable interrelationships—S, R, and Q—will often play the role of data
in this text; that is, they will be the main starting points for applying MVS
methods (with the raw data also remaining relevant in MVS in its own
right). We also emphasize that, in this sense, for a given empirical data set,
the covariance, correlation, and SSCP matrices are only the beginning and
the means rather than the end of MVS applications.
1.6 Illustration Using Statistical Software

In this section, we illustrate the previous discussions using two of the most
widely circulated statistical analysis software, SPSS and SAS. To achieve
this goal, we use data from a sample of n ¼ 32 freshmen in a study of the
relationship between several variables that are defined below. Before we
commence, we note that such relatively small sample size examples will
occasionally be used in this book merely for didactic purposes, and stress
that in empirical research it is strongly recommended to use large samples
whenever possible. The desirability of large sample sizes is a topic that
has received a considerable amount of attention in the literature because it
is well recognized that the larger the sample, the more stable the parameter
estimates will be, although there are no easy to apply general rules for
sample size determination. (This is because the appropriate size of a
sample depends in general on many factors, including psychometric prop-
erties of the variables selected, the strength of relationships among them,
number of observed variables, amount of missing data, and the distribu-
tional characteristics of the analyzed variables; Marcoulides & Saunders,
2006; Muthén & Muthén, 2002). In this example data set, the following
variables are considered: (a) GPA at the beginning of fall semester (called
GPA in the data file ch1ex1.dat available from www.psypress.com=
applied-multivariate-analysis), (b) an initial math ability test (called
INIT_AB in that file), (c) an IQ test score (called IQ in the file), and (d) the
number of hours the person watched television last week (called
HOURS_TV in the file). For ease of discussion, we also assume that there
are no anomalous values in the data on any of the observed variables. In
Chapter 3, we revisit the issue of anomalous values and provide some
guidance concerning how such values can be examined and assessed.
For the purposes of this illustration, we go through several initial data
analysis steps. We begin by studying the frequencies with which scores
occur for each of these variables across the studied sample. To accomplish
this, in SPSS we choose the following menu options (in the order given
next) and then click on the OK button:
Analyze ! Descriptive Statistics ! Frequencies
(Upon opening the data file, or reading in the data to be analyzed, the
‘‘Analyze’’ menu choice is available in the toolbar, with the ‘‘Descriptive
Statistics’’ becoming available when ‘‘Analyze’’ is chosen, and similarly
for ‘‘Frequencies.’’ Once the latter choice is clicked, the user must move
over the variables of interest into the variable selection window.)
To obtain variable frequencies with SAS, the following set of commands
can be used:
DATA CHAPTER1;
INFILE ‘ch1ex1.dat’;
INPUT GPA INIT_AB IQ HOURS_TV;
PROC FREQ;
RUN;
In general, SAS program files normally contain commands that describe

the data to be analyzed (the so-called DATA statements), and the type of
procedures to be performed (the so-called PROC statements). SAS PROC
statements are just like computer programs that perform various manipu-
lations, and then print the results of the analyses (for complete details, see
the latest SAS User’s Guide and related manuals). The INFILE command
statement merely indicates the file name from which the data are to be
read (see also specific software arrangements regarding accessing raw

data), and the following INPUT statement invokes retrieval of the data
from the named file, in the order of specified free-formatted variables.
Another way to inform SAS about the data to use is to include the entire
data set (abbreviated below to the first five observations to conserve
space) in the program file as follows:
DATA CHAPTER1;
INPUT GPA INIT_AB IQ HOURS_TV;
CARDS;
2.66 20 101 9
2.89 22 103 8
3.28 24 99 9
2.92 12 100 8
4 21 121 7
;
PROC FREQ;
RUN;
Once either of these two command sequences is submitted to SAS, two

types of files reporting results are created. One is called the SAS log
file and the other is referred to as the SAS output file. The SAS log file
contains all commands, messages, and information related to the execu-
tion of the program, whereas the SAS output file contains the actual
statistical results.
The outputs created by running the software, SPSS and SAS, as indicated
above are as follows. For ease of presentation, we separate them by pro-
gram and insert clarifying comments after each output section accordingly.
SPSS output notes

Frequencies
Notes
Output Created
Comments
Input Data D:\Teaching\Multivariate.Statistics\
Data\Lecture1.sav
Filter <none>
Weight <none>
Split File <none>
N of Rows in
Working Data File 32
Missing Value Definition of Missing User-defined missing values are
Handling treated as missing.
Cases Used Statistics are based on all cases

with valid data.
Syntax FREQUENCIES
VARIABLES ¼ gpa init_ab iq
hours_tv=ORDER ¼ ANALYSIS.
Resources Elapsed Time 0:00:00.09
Total Values Allowed 149796
SAS output log file
NOTE: Copyright (c) 2002–2003 by SAS Institute Inc.,

Cary, NC, USA.
NOTE: SAS (r) 9.1 (TS1M3)
Licensed to CSU FULLERTON-SYSTEMWIDE-T=R, Site
0039713013.
NOTE: This session is executing on the WIN_PRO platform.
NOTE: SAS 9.1.3 Service Pack 1
NOTE: SAS initialization used:
real time 16.04 seconds
cpu time 1.54 seconds
1 DATA CHAPTER1;
2 INPUT GPA INIT_AB IQ HOURS_TV;
3 CARDS;
NOTE: The data set WORK.CHAPTER1 has 32 observations
and 5 variables.
NOTE: DATA statement used (Total process time):
36 ;
37 PROC FREQ;
38 RUN;
NOTE: There were 32 observations read from the data set
WORK.CHAPTER1.
NOTE: PROCEDURE FREQ used (Total process time):
To save space, in the remainder of the book we dispense with these

beginning output parts for each considered analytic session in both soft-
ware, which echo back the input submitted to the program and=or contain
information about internal arrangements that the latter invokes in order to
meet the analytic requests. We also dispense with the output section titles,
and supply instead appropriate headings and subheadings.
SPSS frequencies output

Statistics
GPA INIT_AB PSI IQ HOURS_TV
N Valid 32 32 32 32 32
Missing 0 0 0 0 0
This section confirms that we are dealing with a complete data set, that is,
one having no missing values, as indicated in Section 1.4.
Frequency Table
GPA
Valid Cumulative
Frequency Percent Percent Percent
Valid 2.06 1 3.1 3.1 3.1

2.39 1 3.1 3.1 6.3
2.63 1 3.1 3.1 9.4
2.66 1 3.1 3.1 12.5
2.67 1 3.1 3.1 15.6
2.74 1 3.1 3.1 18.8
2.75 1 3.1 3.1 21.9
2.76 1 3.1 3.1 25.0
2.83 2 6.3 6.3 31.3
2.86 1 3.1 3.1 34.4
2.87 1 3.1 3.1 37.5
2.89 2 6.3 6.3 43.8
2.92 1 3.1 3.1 46.9
3.03 1 3.1 3.1 50.0
3.10 1 3.1 3.1 53.1
3.12 1 3.1 3.1 56.3
3.16 1 3.1 3.1 59.4
3.26 1 3.1 3.1 62.5
3.28 1 3.1 3.1 65.6
3.32 1 3.1 3.1 68.8
3.39 1 3.1 3.1 71.9
3.51 1 3.1 3.1 75.0
3.53 1 3.1 3.1 78.1
Valid Cumulative
3.54 1 3.1 3.1 81.3

3.57 1 3.1 3.1 84.4
3.62 1 3.1 3.1 87.5
3.65 1 3.1 3.1 90.6
3.92 1 3.1 3.1 93.8
4.00 2 6.3 6.3 100.0
Total 32 100.0 100.0
INIT_AB
Valid Cumulative
Valid 12.0 1 3.1 3.1 3.1

14.0 1 3.1 3.1 6.3
17.0 3 9.4 9.4 15.6
19.0 3 9.4 9.4 25.0
20.0 2 6.3 6.3 31.3
21.0 4 12.5 12.5 43.8
22.0 2 6.3 6.3 50.0
23.0 4 12.5 12.5 62.5
24.0 3 9.4 9.4 71.9
25.0 4 12.5 12.5 84.4
26.0 2 6.3 6.3 90.6
27.0 1 3.1 3.1 93.8
28.0 1 3.1 3.1 96.9
29.0 1 3.1 3.1 100.0
Total 32 100.0 100.0
IQ
Valid Cumulative
Valid 97.00 2 6.3 6.3 6.3

98.00 4 12.5 12.5 18.8
99.00 4 12.5 12.5 31.3
100.00 1 3.1 3.1 34.4
101.00 7 21.9 21.9 56.3
102.00 1 3.1 3.1 59.4
103.00 2 6.3 6.3 65.6
104.00 1 3.1 3.1 68.8
107.00 1 3.1 3.1 71.9
110.00 2 6.3 6.3 78.1
111.00 1 3.1 3.1 81.3
112.00 1 3.1 3.1 84.4
113.00 2 6.3 6.3 90.6
114.00 1 3.1 3.1 93.8
119.00 1 3.1 3.1 96.9
121.00 1 3.1 3.1 100.0
Total 32 100.0 100.0
HOURS_TV
Valid Cumulative
Valid 6.000 4 12.5 12.5 12.5

6.500 2 6.3 6.3 18.8
7.000 6 18.8 18.8 37.5
7.500 1 3.1 3.1 40.6
8.000 10 31.3 31.3 71.9
8.500 2 6.3 6.3 78.1
9.000 6 18.8 18.8 96.9
9.500 1 3.1 3.1 100.0
Total 32 100.0 100.0
SAS frequencies output
Cumulative Cumulative
GPA Frequency Percent Frequency Percent
2.06 1 3.13 1 3.13

2.39 1 3.13 2 6.25
2.63 1 3.13 3 9.38
2.66 1 3.13 4 12.50
2.67 1 3.13 5 15.63
2.74 1 3.13 6 18.75
2.75 1 3.13 7 21.88
2.76 1 3.13 8 25.00
2.83 2 6.25 10 31.25
2.86 1 3.13 11 34.38
2.87 1 3.13 12 37.50
2.89 2 6.25 14 43.75
2.92 1 3.13 15 46.88
3.03 1 3.13 16 50.00
3.1 1 3.13 17 53.13
3.12 1 3.13 18 56.25
3.16 1 3.13 19 59.38
3.26 1 3.13 20 62.50
3.28 1 3.13 21 65.63
3.32 1 3.13 22 68.75
3.39 1 3.13 23 71.88
3.51 1 3.13 24 75.00
3.53 1 3.13 25 78.13
3.54 1 3.13 26 81.25
3.57 1 3.13 27 84.38
3.62 1 3.13 28 87.50
3.65 1 3.13 29 90.63
3.92 1 3.13 30 93.75
4 2 6.25 32 100.00
INIT_AB Frequency Percent Frequency Percent
12 1 3.13 1 3.13
14 1 3.13 2 6.25
17 3 9.38 5 15.63
19 3 9.38 8 25.00
20 2 6.25 10 31.25
21 4 12.50 14 43.75
22 2 6.25 16 50.00
23 4 12.50 20 62.50
24 3 9.38 23 71.88
25 4 12.50 27 84.38
26 2 6.25 29 90.63
27 1 3.13 30 93.75
28 1 3.13 31 96.88
29 1 3.13 32 100.00
IQ Frequency Percent Frequency Percent
97 2 6.25 2 6.25
98 4 12.50 6 18.75
99 4 12.50 10 31.25
100 1 3.13 11 34.38
101 7 21.88 18 56.25
102 1 3.13 19 59.38
103 2 6.25 21 65.63
104 1 3.13 22 68.75
107 1 3.13 23 71.88
110 2 6.25 25 78.13
111 1 3.13 26 81.25
112 1 3.13 27 84.38
113 2 6.25 29 90.63
114 1 3.13 30 93.75
119 1 3.13 31 96.88
121 1 3.13 32 100.00
HOURS_TV Frequency Percent Frequency Percent
6 4 12.50 4 12.50
6.5 2 6.25 6 18.75
7 6 18.75 12 37.50
7.5 1 3.13 13 40.63
8 10 31.25 23 71.88
8.5 2 6.25 25 78.13
9 6 18.75 31 96.88
9.5 1 3.13 100 100.00
These SPSS and SAS frequency tables provide important information

regarding the distribution (specifically, the frequencies) of values that each
of the variables is taking in the studied sample. In this sense, these output
sections also tell us what the data actually are on each variable. In par-
ticular, we see that all variables may be considered as (approximately)
continuous.
The next step is to examine what the descriptive statistics are for every
measure used. From these statistics, we learn much more about the stud-
ied variables, namely their means, range of values taken, and standard
deviations. To this end, in SPSS we use the following sequence of menu
option selections:
Analyze ! Descriptive statistics ! Descriptives
In SAS, the procedure PROC MEANS would need to be used and stated
instead of PROC FREQ (or in addition to the latter) in the second-to-last
line of either SAS command file presented above.
Each of the corresponding software command sets provides identical
output (up to roundoff error) shown below.
SPSS output
Descriptive Statistics
N Minimum Maximum Mean Std. Deviation
GPA 32 2.06 4.00 3.1172 .46671

INIT_AB 32 12.0 29.0 21.938 3.9015
IQ 32 97.00 121.00 104.0938 6.71234
HOURS_TV 32 6.000 9.500 7.71875 1.031265
Valid N (listwise) 32
SAS output
Variable N Mean Std Dev Minimum Maximum
GPA 32 3.1171875 0.4667128 2.0600000 4.0000000

INIT_AB 32 21.9375000 3.9015092 12.0000000 29.0000000
IQ 32 104.0937500 6.7123352 97.0000000 121.0000000
HOURS_TV 32 7.7187500 1.0312653 6.0000000 9.5000000
To obtain standardized measures of variable interrelationships, we

produce their correlation matrix. In SAS, this is accomplished by including
the procedure PROC CORR at the end of the program file used previously
(or instead of PROC FREQ or PROC MEANS), whereas in SPSS this is

accomplished by requesting the following series of menu options:
Analyze ! Correlations ! Bivariate
These SPSS and SAS commands furnish the following output:
SPSS output
Correlations
GPA INIT_AB IQ HOURS_TV
GPA Pearson Correlation 1 .387* .659** .453**

Sig. (2-tailed) . .029 .000 .009
N 32 32 32 32
INIT_AB Pearson Correlation .387* 1 .206 .125

Sig. (2-tailed) .029 . .258 .496
N 32 32 32 32
IQ Pearson Correlation .659** .206 1 .784**

Sig. (2-tailed) .000 .258 . .000
N 32 32 32 32
HOURS_TV Pearson Correlation .453** .125 .784** 1

Sig. (2-tailed) .009 .496 .000 .
N 32 32 32 32
*. Correlation is significant at the 0.05 level (2-tailed).
**. Correlation is significant at the 0.01 level (2-tailed).
SAS output
Pearson Correlation Coefficients, N ¼ 32

Prob > jrj under H0: Rho ¼ 0
GPA 1.00000 0.38699 0.65910 0.45276
0.0287 <.0001 0.0093
INIT_AB 0.38699 1.00000 0.20594 0.12477
0.0287 0.2581 0.4963>
IQ 0.65910 0.20594 1.00000 0.78362
<.0001 0.2581 <.0001>
HOURS_TV 0.45276 0.12477 0.78362 1.00000
0.0093 0.4963 <.0001
For the purposes of this introductory chapter, we will not place

emphasis on the p-values that are associated with each of the numerical
cells with estimated correlations in these tables (labeled in the SPSS output
‘‘Sig. (2-tailed)’’ and in the SAS output ‘‘Prob > jrj under H0: Rho ¼ 0’’).
Instead, we primarily look at these correlations as descriptive measures of
linear variable interrelationships, and thereby interpret mainly their sign
and magnitude (i.e., distance from 1 or 1, or closeness to 0; see also
earlier discussion in Section 1.4.).
In order to generate a covariance matrix and an SSCP matrix for the
studied variables in SAS, instead of (or in addition to) PROC FREQ in
the second-last line of either of the above SAS command files we use the
procedure PROC CORR with its options ‘‘COV’’ and ‘‘SSCP,’’ whereas in
SPSS the following set of menu options can be used:
Analyze ! Correlate ! Bivariate (click Options and then check cross

products and deviations)
These commands furnish the following output.
SAS output
SSCP Matrix
GPA 317.6919 2210.1100 10447.3600 763.1900

INIT_AB 2210.1100 15872.0000 73241.0000 5403.0000
IQ 10447.3600 73241.0000 348133.0000 25543.0000
HOURS_TV 763.1900 5403.0000 25543.0000 1939.5000
Covariance Matrix, DF ¼ 31
GPA 0.21782087 0.70465726 2.06478831 0.21791331

INIT_AB 0.70465726 15.22177419 5.39314516 0.50201613
IQ 2.06478831 5.39314516 45.05544355 5.42439516
HOURS_TV 0.21791331 0.50201613 5.42439516 1.06350806
SPSS output
Correlations
GPA INIT_AB PSI IQ HOURS_TV
GPA Pearson Correlation 1 .387* .040 .659** .453**

Sig. (2-tailed) . .029 .829 .000 .009
Sum of Squares 6.752 21.844 .289 64.008 6.755
and Cross-products
Covariance .218 .705 .009 2.065 .218
N 32 32 32 32 32
INIT_AB Pearson Correlation .387* 1 .113 .206 .125

Sig. (2-tailed) .029 . .539 .258 .496
Sum of Squares 21.844 471.875 6.875 167.187 15.562
and Cross-products
Covariance .705 15.222 .222 5.393 .502
N 32 32 32 32 32
PSI Pearson Correlation .040 .113 1 .035 .097

Sig. (2-tailed) .829 .539 . .848 .598
Sum of Squares .289 6.875 7.875 3.688 1.563
and Cross-products
Covariance .009 .222 .254 .119 .050
N 32 32 32 32 32
IQ Pearson Correlation .659** .206 .035 1 .784**

Sig. (2-tailed) .000 .258 .848 . .000
Sum of Squares 64.008 167.187 3.688 1396.719 168.156
and Cross-products
Covariance 2.065 5.393 .119 45.055 5.424
N 32 32 32 32 32
HOURS_TV Pearson Correlation .453** .125 .097 .784** 1

Sig. (2-tailed) .009 .496 .598 .000 .
Sum of Squares 6.755 15.562 1.563 168.156 32.969
and Cross-products
Covariance .218 .502 .050 5.424 1.064
N 32 32 32 32 32
*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).
On the basis of these tables, we can easily extract the needed empirical
covariance matrix S. For example, using the output provided by SPSS, we
look into each of the five panels pertaining to the analyzed variable, and
find the fourth row (titled ‘‘Covariance’’). Note that the respective diag-
onal elements are actually the squares of the entries in the column ‘‘Std.
Deviations’’ of the output table obtained earlier in this section using the
‘‘Descriptives’’ procedure. We thus furnish the following sample covar-
iance matrix S for the study under consideration:
2 3
:218
6 :705 15:222 7
6 7
S¼6
6 :009 :222 :254 7:
7 (1:14)
4 2:065 5:393 :119 45:055 5
:218 :502 :050 5:424 1:064
Finally, the above SPSS and SAS output tables can also be used to extract
the SSCP matrix Q. For example, using the last considered SPSS output,
we look into each of the five panels pertaining to the analyzed variables
and extract its third row (titled ‘‘Sum of Squares and Cross Products’’). It
is important to note, however, that in the output provided by SAS it is the
raw data SSCP matrix U that is rendered, whereas SPSS provides directly
the (deviation) SSCP matrix Q. In order to obtain the deviation SSCP
matrix, Q, using the output provided by SAS, one would have to compute
the difference between each entry of the SSCP matrix for the raw data
displayed in the SAS output, and the corresponding sum of squared mean
or mean products (counted once for each subject). For example, with the
value determined by SAS for the variable GPA (i.e., 317.6919), via sub-
traction of 32 times its squared mean (i.e., the squared mean 3.1171875
summed 32 times, resulting in 310.93945), one obtains the value 6.752,
which is the one displayed as the corresponding SSCP matrix element in
the SPSS output. In this way, either from the SPSS or SAS output, we can
obtain the following deviation SSCP matrix for the five variables under
consideration:
2 3
6:752
6 21:844 471:875 7
6 7
Q¼6
6 :289 6:875 7:875 7:
7 (1:15)
4 64:008 167:187 3:688 1396:719 5
6:755 15:562 1:563 168:156 32:969
Note that as mentioned earlier, the elements of Q are much larger than the
respective ones of S—the reason is that the former are 31 times (i.e., n 1
times in general) those of the latter.
We conclude this chapter with the following general cautionary remark.
In applied research in the behavioral, social, and educational sciences,
often all elements of a correlation matrix are evaluated for significance.
However, if their number is large, even if in the population all correlation
coefficients (between all pairs of variables) are 0, it is likely that some will
turn out to be significant purely by chance. Therefore, when faced with a
correlation matrix a researcher should better not evaluate more than a
limited number of correlations for significance, unless he or she has an
a priori idea exactly which very few of them to examine.
2
Elements of Matrix Theory
Multivariate statistical methods have been developed to deal with the

inherent complexity of studies based on multiple interrelated variables.
To handle this complexity, and still keep clarity about the analytic ideas
underlying the methods, it is essential that one makes use of systematic
means that aim at reducing matters to more manageable levels. Many
such means are provided by matrix theory, and a discussion of some of its
elements is the subject of this chapter. In particular, we introduce the
reader to the compact language of matrices, and familiarize them with a
number of basic matrix operations, including the important notions of
determinant and trace that will be of special relevance for later develop-
ments in the book. In addition, we define the concept of distance and
particularly the so-called Mahalanobis distance (MD), which will play an
instrumental role in the next chapters.
2.1 Matrix Definition

Matrix theory (MT) is a compact language that we will use in this text in
part as a means to underscore important aspects of multivariate statistics
(MVS). MT readily enables one to understand a number of multivariate
techniques using their univariate analogs and counterparts. In addition,
MT allows us to handle efficiently the complexity of MVS in settings with
multiple observed variables.
A matrix is in general a rectangular set, a table, or an array of numbers.
These often, but not always, are in this book the scores of subjects on some
set of observed variables—e.g., a set of intelligence tests—whereby sub-
jects are represented by their rows and variables by their columns. For
most of the text, matrices will (a) consist of the data collected in a study
(i.e., will have the form of a data matrix—see the pertinent discussion in
the preceding chapter), or (b) represent indices of interrelationships
between studied variables, such as the covariance, correlation, and sum-
of-squares and cross-products (SSCP) matrices S, R, and Q from Chapter 1.
For example, the data matrix provided below (also described previously
31
as Table 1.1 in the last chapter, apart from its first column) would repre-
sent a 4 3 6 matrix (read ‘‘four by six matrix’’):
2 3
45 55 47 1 3 33
6 51 54 57 0 1 23 7
X ¼6
4 40
7:
51 46 1 2 43 5
49 45 48 0 3 42
Other examples of matrices considered in Chapter 1 included those

representing the matrix S in Equation 1.2, which was a 5 3 5 matrix; the
matrix R in Equation 1.8 that was a 4 3 4 matrix; and the matrix Q in
Equation 1.3, which was of size 3 3 3. We will use capital letters to denote
matrices in this text.
Being defined as a two-dimensional entity means that any matrix has
size. This size is determined by the number of its rows (say r, with r 1) as
well as that of its columns (say c, with c 1), and we denote it as r 3 c.
There are no restrictions on r and c, as long as they are whole positive
numbers. In particular, as indicated above when considering the correl-
ation, covariance and SSCP matrices, r and c can (but need not) be equal to
one another, in which case one is dealing with a square matrix. Further-
more, the case in which r ¼ 1 is also possible, when the matrix is called a
row vector. Similarly, the case of c ¼ 1 is possible, when the matrix is called
a column vector. In fact, both r ¼ 1 and c ¼ 1 are possible, and thus any
number can be viewed as a 1 3 1 matrix, also called scalar.
For example, taking only the second row of the matrix X above, we
could represent it as the following 1 3 6 row vector
x ¼ [51, 54, 57, 0, 1, 23], (2:1)
where we place the elements of the vector within brackets, a convention

followed throughout the book; similarly, we will underline the lowercase
symbol used to denote the vector, as in the left-hand side of Equation 2.1.
Alternatively, the data of the four subjects displayed in the second
column of the above matrix X (i.e., their values obtained on the Reading
test; see Chapter 1 and Table 1.1) would represent a 4 3 1 column vector:
2
3
55
6 54 7
y¼6 7
4 51 5: (2:2)
45
In the methodological literature, a commonly accepted rule is to assume

a column vector when referring to a vector; otherwise the reference ‘‘row
vector’’ is explicitly used.
Elements of Matrix Theory 33
2.2 Matrix Operations, Determinant, and Trace

In order to use matrix theory effectively, certain operations need to be
carried out on matrices. Below we provide a discussion of some essential
operations needed to enhance one’s understanding of MVS. We begin
with the notion of matrix equality.
Matrix equality. Two (or more) matrices are said to be equal if (a) they are
of the same size, and (b) each element (i.e., each entry) of one of them is
identical to the element in the other, that is, in exactly the same position.
That is, equal matrices have identical corresponding elements. For
example, if we define the matrix A as
2 3
2 3
A ¼ 43 25 (2:3)
4 2
and the matrix B as

2 3
2 3
B ¼ 4 3 2 5, (2:4)
4 2
then these are two equal matrices, which is commonly written as A ¼ B.

On the other hand, if we consider the matrix C
2 3
2 3
C ¼ 42 3 5, (2:5)
4 2
then A does not equal C. That is, even though the matrices A and C consist
of the same elements (numbers in this case), they are positioned at different
places, and for this reason A is not equal to C, which is written as A 6¼ C.
Throughout this book, we will use a special notation for a matrix with
real numbers as its elements—specifically, we will place the general mat-
rix element within brackets. For example, the matrix
2 3
a11 a12 . . . a1c
6 a21 a22 . . . a2c 7
A¼6
4 :
7, (2:6)
: : : 5
ar1 ar2 . . . arc
where aij are some real numbers, will be symbolized in this notation as
A ¼ [aij] (i ¼ 1, 2, . . . , r; j ¼ 1, 2, . . . , c), with the last statement in parentheses
indicating the range which i and j cover. (In the literature, this statement is
sometimes omitted, if the range is clear from the context or not of rele-
vance.) Thus, the matrices A ¼ [aij] and B ¼ [bij] that are of the same
size will be equal if and only if aij ¼ bij for all i and j; otherwise, if for
some i and j it is the case that aij 6¼ bij, then A 6¼ B (i ¼ 1, 2, . . . , r; j ¼ 1, 2, . . . , c).
Now that we have defined when two matrices are equal, we can move
on to a discussion of matrix operations.
Matrix addition and subtraction. These operations are defined only for
matrices that are compatible (also sometimes referred to as conformable
or simply conform), i.e., matrices that are of the same size. Under such
conditions, the sum and the difference of two matrices is obtained by
simply adding or subtracting one by one their corresponding elements.
That is, if C ¼ [cij] is the sum of the conform matrices A ¼ [aij] and B ¼ [bij],
in other words if C ¼ A þ B, then cij ¼ aij þ bij. Similarly, if D ¼ [dij]
denotes the difference between the matrices A and B, that is D ¼ A B,
then dij ¼ aij bij (i ¼ 1, 2, . . . , r; j ¼ 1, 2, . . . , c). For example, if A and B
are the first and second matrix in the middle part of the following
Equation 2.7, respectively, their sum and difference are readily obtained
as follows:
2 3 2 3 2 3
3 7 4 11 13 14 14 20 18
C ¼ A þ B ¼ 42 4 55 þ 4 2 4 5 5¼4 4 8 10 5 (2:7)
1 4 3 2 3 4 3 7 7
and
2 3 2 3 2 3
3 7 4 11 13 14 8 6 10
D ¼ A B ¼ 42 4 55 4 2 4 5 5¼4 0 0 0 5: (2:8)
1 4 3 2 3 4 1 1 1
Addition and subtraction operations of conform matrices are straightfor-

wardly extended to any number of matrices. We also note that matrices do
not need to be square, as they happen to be in the last two examples, in
order for their sum and difference to be defined (which is readily seen from
their general definition given above), as long as they are of the same size.
Matrix multiplication with a number (scalar). A matrix is multiplied with a
number (scalar) by multiplying each element of the matrix with that
number. That is, if the matrix A ¼ [aij] is considered, then gA ¼ [gaij],
where g is a scalar (i ¼ 1, 2, . . . , r; j ¼ 1, 2, . . . , c). For example, if g ¼ 3, and
2 3
2 4
A ¼ 43 2 5,
1 5
Exploring the Variety of Random
Documents with Different Content
"Dear me!" said his grandfather. "And on what, pray?"
"Well, first I thought," said Anne-Hilarion, "that I would give all

the money in my money-box to M. le Lieutenant Tollemache for
saving Papa, for since M. Tollemache is not poor, like M. de Soucy, it
is permitted to offer him money—is it not? But Papa said . . . What
was it you said, Papa?"
The Marquis smiled at his small and earnest son, and put his arm
round him. "I believe I told you to keep it for yourself, Anne."
"But I did not save you, Papa!" exclaimed the child, almost
indignantly.
René de Flavigny's eyes sought the fire. "I would not be too sure
of that," he said. "On whose account, do you suppose, Anne, did Mr.
Tollemache take all that trouble and risk for me?"
"I suppose," replied the little boy, wrinkling his forehead, "for St.
Michel, because I asked him very particularly to take care of you."
"Yes," repeated René, "as I say, it was you who saved me, my
son. But not, perhaps, quite in the way you think," he added to
himself.
There was a moment's pause, during which Anne apparently

resolved not to pursue this question, for he went on with a business-
like air: "I have now quite resolved what I will do with my money,
which is now a great sum, with what Grandpapa gave me at
Christmas. I shall not give it to M. le Lieutenant."
"Well?" queried Mr. Elphinstone, looking at him over his

spectacles. "This suspense is very hard to bear, Anne."
"I shall spend it on going to Portsmouth to see M. le Chevalier."

The two men looked at each other at this announcement. "What
next?" asked the Marquis, amused.
"After that I shall begin to save for the new box," responded his
son, taking the inquiry literally. "For though to go to Portsmouth will
not cost as much as going to France would have done, I expect it
will quite empty the old one."
"And a very good thing too," remarked his grandfather, "if you
are going to employ your savings to such ends. We have had
enough, in this house, of your jaunts, my bairn."
"But it was you, Grandpapa, that sent me to Canterbury!" said

Anne, turning an accusing gaze on the old gentleman. Mr.
Elphinstone collapsed.
"True, only too true!" he murmured. "But, child, your father is

going down to Portsmouth to see M. de la Vireville directly he is able
to travel. He has already written to him to that effect."
"But that will be quite a long time yet, I know," returned the
Comte wisely. "I heard Dr. Collins say so."
"You could write M. de la Vireville a letter," suggested his

grandfather.
"But I want to see him!" repeated Anne. "One does not see a
person by writing him a letter."
"This child's arguments are difficult to controvert," remarked Mr.

Elphinstone to his son-in-law. "I do not see any reply to that."
"Except perhaps this," suggested the Marquis. "Are you sure,

Anne, that on his side M. de la Vireville wants to see you just now?
He is rather ill, you know."
Anne-Hilarion gave this due consideration. "But if I were ill, Papa,
I should want to see you and Grandpapa. It would make me feel
better—as when I had whooping-cough last year."
"And you think that your presence would have a similar good
effect on M. de la Vireville? You are not wanting in assurance, my
son!"
Anne smiled, because he knew that he was being teased, and,

the clock striking at that moment, he slipped out of his father's arm.
"Will you please to think about it, Papa, while I have my walk?" he
said coaxingly.
After he had gone Mr. Elphinstone turned over his manuscripts

for a minute or two. Then he looked across at his son-in-law, who
was staring again into the fire.
"I could take the child to Portsmouth, René, if you wish him to go
—and can trust him to me," he said. "I do not know what you feel,
but it seems to me that it might be some slight attempt to repay
that great debt which we owe on Anne's behalf—and M. de la
Vireville was so fond of the child that he might really be glad to see
him."
René de Flavigny looked up and smiled. "How well you read my

thoughts, sir!"
(2)
On that same remarkably sunny day in late January the old
Bishop, in a long black cloak, was walking up and down the little
walled garden at Portsmouth under a sky as blue as May's. The
forerunners of spring had arrived, and the sight of that vanguard
evidently gave him a lively pleasure. He was standing looking at the
border when he heard a step, and observed Mme. de la Vireville
approaching him. She had come to the house earlier in the day, but
he had not seen her.
"It is almost spring already, Madame," he remarked to her. "Look

at that patch of aconites!"
Mme. de la Vireville did not obey him. She came up, kissed his
ring, and said with the directness of a child, "It is not spring in my
heart, Monseigneur. Your Grandeur knows why."
The Bishop may have had the eyes of a mystic, but they were by
no means blind to mundane affairs. He looked at her now. "Yes, I
know, my daughter. I have been wishing for some time to speak
with you of this. You will not feel cold if we walk up and down a little
in the sun?"
She shook her head and turned with him. At their feet the
snowdrops stood smiling and shivering behind little rows of box. "I
have just come down from Fortuné's room, Monseigneur. He is no
better, this morning—not so well, I think."
They took a turn in silence. "Forgive me if I am impertinent," said

the Bishop, rather suddenly. "I have been wondering of late why
your son has never married. How old is he—forty?"
Mme. de la Vireville shook her head with a sad little smile. "Only
thirty-five, Monseigneur. As for his marrying, I have long greatly
desired it, but he will not look at a woman. He has good reason,
perhaps." She hesitated, then went on. "There was one, ten years
ago . . . he loved her only too well. She too seemed to love him
dearly, and became his affianced wife. On the very day before their
marriage she fled from her home with another man, whom she had
only known for a week or two. That man was Fortuné's intimate
friend."
"And then?" asked the Bishop.

"Fortuné called him out—he could hardly do less. The scar which
you may have remarked on his face, Monseigneur, is a memorial of
their encounter. It is where his false friend's bullet wounded him—he
can never look in a glass without seeing that reminder. They used
pistols, not swords—I do not know why—and drew lots for the order
of firing. And though my son, since he fired second, had this man
who had so deeply injured him absolutely at his mercy, though he
was half beside himself with grief and rage, he spared him, for her
sake, and fired in the air."
"That was well done," said the Bishop.
Mme. de la Vireville laughed. "Was it not, Monseigneur! It was

not easily done, either, that I know. Can you guess what Fortuné's
reward was? After a year she left this man, to whom she was not
wedded, and married another."
The Bishop looked very grave. "And your son, Madame, after so
bitter a betrayal, has conceived a hatred of all women?"
"Hardly that, Monseigneur. It is more hopeless even than that—

for such an aversion might change. No, I am almost sure that
against his will he loves her still. That is the tragedy."
"She is still living—her husband also?"
"I believe so."
"Perhaps M. de la Vireville hopes to marry her in the end, if—as

may so easily happen in these sad days—she should be suddenly left
a widow?"
"No, Monseigneur, he would never do that. He has never forgiven

her. But he will not look at another woman. I think it would make no
difference to him if he were to hear of her death to-morrow. For him
she has long been dead . . . and yet she is alive. Would God she
were not! Her lover was killed by his side at Quiberon; he told me
the other day." She paused a moment, looking into the distance, and
resumed, with a little gesture: "Do not imagine, Monseigneur, that
Fortuné is always thinking of this. He is not a dreamer; he has
always been a man of action, and a reckless one at that. It is but a
scar now, I think, not a wound—but it is a scar with poison in it. And
over there in Jersey, when I saw him with the little boy . . ." She
stopped, and the tears came into her eyes. "Monseigneur, I believe
that in his heart of hearts Fortuné desires as much to have a son as
I desire to see him with one."
"But," said the Bishop, "there is nothing to prevent his marrying

some day, if he could cut himself loose from this memory. If he could
so cut himself loose, the rest—you must pardon an old man, my
daughter—the rest would not be difficult, would it?"
"Monseigneur, a man who will not look at women is always

attractive to them."
The Bishop smiled. "I suppose that is true. Now would you be
astounded to learn that, before you came, he used sometimes, in
sleep or delirium, to repeat a woman's name? I suppose it was hers
who betrayed him."
"I do not think that likely, Monseigneur," said Mme. de la Vireville.

"He has not mentioned her name for years. And that it should have
been any other woman's is impossible."
"Then perhaps my ears deceived me," replied the Bishop, looking

as if he were pretty sure that they had not. "In that case I shall
perhaps not be indiscreet if I tell you the name—admitting frankly
that some of the context puzzled me. It was—'Anne.'"
It may be seen what bond of error united the old French Bishop
and the middy of the Pomone.
Mme. de la Vireville clasped her little hands together. "But,

Monseigneur, that exactly bears out what I said about his desiring a
son. Anne is the name of the little boy I was referring to just now—
Anne-Hilarion de Flavigny, his friend's son—the friend about whose
fate, as Your Grandeur knows, Fortuné was anxious, but who proves,
after all, to have been saved at Quiberon. Fortuné had promised the
Marquis de Flavigny to look after the child if he—the Marquis—were
killed."
"But now, as the Marquis escaped, he will not be called upon to

undertake this charge?"
"No, Monseigneur."
"That is a pity," said the Bishop, looking down reflectively at the

radiant face of a little beruffed aconite at his feet. "There are all
sorts of doors which only a child's hands can unlock." And, still
looking at the aconite, he went on gently: "Madame, I should be
doing wrong were I to disguise from you that the doctor does not
think well of the lethargy which seems latterly to have taken
possession of your son, and which appears to have so much
connection with his physical condition."
"I know it," said the poor mother, all the delicate colour gone
from her cheeks. "But what more can I do, Monseigneur? I know
that Fortuné loves me dearly, but I am old, and represent the past to
him, not the future, and it is the past that he needs to forget. . . .
He is ill, it is true—he has been very ill—but never have I seen him
like this. Always, in whatever vicissitudes—and he has been severely
wounded before, and I nursed him in Jersey—always he has been
full of gaiety and courage. Now all that seems to have deserted him,
as if he did not care to live."
"Madame, is that, after all, so much to be wondered at?" asked

the Bishop gravely. "If you or I had fought at Quiberon, and had
seen nearly all our comrades massacred in cold blood, might we not
be tempted to feel the same? There is much buried on that shore
which engulfed so many hopes. I think M. de la Vireville has left his
there, as others their lives. There is not, I fancy, any great difference
between the two losses. . . . Still, as I said, a child's hand holds
many keys, to shut or to open." He stooped at last, a little painfully,
and picked the aconite, and added to himself, "As we say to the
Child who was Himself the Key . . . O Clavis David, qui aperis et
nemo claudit; claudis et nemo aperit, veni et educ vinctum de domo
carceris, sedentem in umbra mortis. . . . I wish your son could have
had the care of this child you speak of."
Mme. de la Vireville could not reply. She had hidden her face in
her hands, and the tears were trickling through them. The little old
man, holding the golden flower in his fingers, stood and looked at
her with a great pity in his eyes.
Suddenly, however, something else came into them—a gleam of

recollection. He looked half doubtfully upon the weeping woman
before him, compressed his lips, then appeared to make up his
mind.
"My daughter," he said, "it has only just come back to my

memory, strangely enough, that one night . . . now this, I fear, really
is betraying an involuntary confidence, but for your sake I am going
to do it . . . one night I heard your son murmuring to himself a
name which can only have been a woman's. But perhaps, again, it
was hers. . . ."
Mme. de la Vireville raised her tear-stained face from her hands.

"What was the name, Monseigneur?"
"'Raymonde,'" said the Bishop.
It was no coup manqué this time. The little mother gazed at him
thunderstruck, amazement, incredulity, and something that might
almost have been a strangled joy chasing each other over her fragile
countenance.
"'Raymonde'?" she repeated. "I . . . it cannot be . . . I know no
one of that name!"
"But evidently your son does, Madame," suggested His Grandeur,

unable to restrain the phantom of a smile. "It was the only time I
ever heard him mention it. He seemed to be beseeching this
'Raymonde' to come to him."
Mme. de la Vireville had no words. Nor had she tears now; her
astonishment had dried them. She stared at Monseigneur, who stood
there with the bright aconite flower in his pale old hands, which
were folded across the purple sash showing between the folds of his
cloak, and she said nothing.
"Your experience of the world, my daughter," went on the Bishop,

"must have taught you that even the most devoted son does not
always confide everything to his mother. In this case, doubtless, the
time was not ripe."
The time, however, did seem to him ripe to leave this mother to
reflect on the information that he had just given her, and, the sound
of a clock striking noon issuing most appositely at this moment from
the house, he seized the opportunity to add:
"If you will excuse me, Madame, for a few moments? I must say
my office."
And pulling out his shabby breviary he went off down the path in
a manner more than diplomatic, for he had said Sext before ever
Mme. de la Vireville came into the garden. However, one can always
get ahead with advantage.
But when a conviction of ten years' growth—one, moreover,

which you have just been stoutly affirming with your own lips—is as
suddenly felled as was Mme. de la Vireville's about her son, it is
natural to find its collapse somewhat devastating. Fortuné's mother,
hardly aware that Monseigneur had left her, stood beside the
snowdrops, certainly more engrossed than was Monseigneur himself
at the other end of the path—and the antiphon to her Hours was a
name she had never heard before.
CHAPTER XXXVII
The Child unlocks the Door
(1)
The Chevalier de la Vireville was lying, as he had so often lain,
staring at the stuffed trout and the triptych of the Assumption. The
adventurous and disappointed spirit which dwelt in him, that had
always faced with a jest every blow of circumstance—but one—had
indeed ebbed very low. Although the sun which five days ago had
opened the aconites in the garden came streaming into the room in
a way to rejoice the heart of the room's late owner, who had given it
up for that very reason, the present occupant was perfectly
indifferent as to whether the sun shone or no. He was back at
Quiberon in the rain—back at Auray in the little chapel, emptied of
all its victims save him alone—back in front of those levelled muskets
which he alone had escaped. . . . Why, in God's name, had fate so
marked him out for safety? Why had he put himself to the agony of
that derelict voyage to Houat and the long suffering that came after?
Mostly because of his promise to his friend, and because he thought
Anne wanted him . . . or because he wanted Anne. But he was not
needed, neither by the boy nor by anyone. He had built a foolish
dream on the assumption of René's death, which to him had seemed
a certainty. The dream had proved baseless, like everything else,
and all it had done was to plant in his sincere thankfulness for
René's escape a thorn of regret which was horrible to him, and
made him, who had suffered so bitterly from treachery, feel a traitor
himself.
No, he was not needed any more; he had done his work. Or
rather, he had not even that consolation, for everything to which he
had set his hand at Quiberon had been a failure, even though the
fault were not his. His men, whom he had never once led to victory
in the Morbihan, were dead or disbanded. There was nothing for him
to do henceforward, and even had there been, he was useless—a
tool that had never been of much account, and was now blunted for
the rest of time.
And further back still. . . . Had he not failed to keep the woman
he loved, the woman who had spurned even the difficult sacrifice he
had made for her sake, when he had spared her lover? He had no
daily perils now to anodyne that ache. But he had another ache to
set against it. . . .
Yes, for ironically enough, at the very hour, five days ago, when
his mother was protesting in the garden to the Bishop that he could
never think of another woman, he was thinking of one. Equally had
Mme. de la Vireville, almost incredulous, carried about for five days
her half-knowledge, longing for her son to speak that name which
meant nothing to her, yet hinted at so much; studying him at odd
moments, trying not to feel hurt at his reserve, puzzled, hoping,
fearing, and far from guessing that this 'Raymonde,' to Fortuné's
mind, was not a subject of which he could ever speak to anyone. For
the thought of her warred against that perverted faithfulness to the
faithless which had made his torment these ten years.
Yet day after day the vision of Mme. de Guéfontaine had been
dwelling more and more constantly with him, and her image, by a
volition, so it seemed, of its own, seeking to efface that other. From
its beginning in that night of agony and effort at Quiberon, its
renewal in the drifting boat, the process had gone on gaining vitality
with time. At Houat he had wakened once from fever to fancy that
he was lying with his head pillowed on her arm, as he had done that
evening when he had stumbled exhausted into the old manoir. That
had proved to be but fever also; his aching head had no such
support, and the English corporal of marines who sat by him, though
kind enough, had very little suggestion of Raymonde about him. Yet
the illusion had given the wounded man such extraordinary pleasure,
and so surprising a sense of rest and security, that he used, in those
barren days, to try to repeat it by a conscious act of the imagination.
She had sat by his couch, once, through the night . . . she had
walked by his horse's stirrup on a spring evening towards the sea . .
. and among the nodding sea-holly she had kissed his hand. It was
his last memory of her, almost as startling as his first. . . .
And now in England he thought of her too—fitfully at first, then

incessantly. But this had served in no way to lighten his depression.
For he was not in love with her, he told himself—how could he be?
Was not all his heart seared over with a fatal memory? Those
shackles could not be loosed now—and even were they miraculously
to be smitten from him, what had he to offer another woman? A
maimed body, an empty purse, a ruined home. . . .
And yet oddly, persistently, he would see himself standing with

her under the larches in front of a house like Kerdronan, that was
perished, and with them stood a little boy like Anne, who did not
need him and was gone from him. . . . He was suddenly possessed
now by that foolish and torturing vision, and lay there clenching and
unclenching his hand, as though in physical pain. No, he and
Kerdronan would go into the dust together, and it was no use
reflecting on what might have been. They were both broken and
done with, he and his home—and no great loss, doubtless, after all.
"Good God!" he exclaimed at last, "what a cowardly fool I am,

lying here and moaning like a sick girl because I am short of an
arm!"
He shut his eyes on that self-condemnation (which had not

helped him), and did not trouble to open them when there came a
knock at the door. Nor, as he still kept them closed when he said
"Come in!" did he see who opened it, and Mr. Elphinstone's face in
the doorway looking at him with a smile that died away to concern.
He only heard the door shut again, and supposed that the visitor, his
mother, or Monseigneur, had decided after all not to disturb him.
(2)
The treble voice, therefore, that said his name suddenly and
softly gave him a violent start. He opened his eyes to see Anne-
Hilarion standing by the closed door, carrying in both his hands a
large glass bowl wherein there swam an enormously magnified
goldfish.
"Anne!" he exclaimed, in a voice of utter incredulity.
And then the sight of him, unchanged, solemn-eyed and

engaging as ever, the touching absurdity of his bringing a goldfish all
the way from London to cheer a sick Chouan, caused La Vireville to
break into a weak laugh that was half something else.
"Oh, M. le Chevalier!" cried Anne, gazing at him. Then he

deposited his precious burden with haste on the floor, and, running
to the bed, flung himself into the welcome of La Vireville's arm.
"My cabbage, my little comrade!" murmured the émigré, and he

kissed the cold, fresh cheek again and again. "You are not changed
at all—yes, I suppose you have grown. . . . Then you have not
forgotten me after all? Have you come all this way to see your poor
bedridden uncle—not by yourself, though, I trust?"
"Oh no," replied Anne-Hilarion, his arms round Fortuné's neck.

"Grandpapa brought me. I wanted to see you so much!" He hugged
him hard, then, drawing back a little, eyed him with a sudden doubt.
La Vireville hastily withdrew his arm and pulled the bedclothes over
his left side.
"Come and get up on the bed," he suggested, "and we can talk

better."
The Comte de Flavigny, needing no second invitation,

incontinently scrambled up—not without difficulty, for the bed was
high.
"I am not too heavy?" he inquired rather anxiously, as he took a
seat on La Vireville's legs. "Papa says I am getting too heavy for
him."
And, as a matter of fact, he had planted himself exactly on one

of the more painful souvenirs of the Ile de Houat; but La Vireville
would not for worlds have asked him to move.
"You are a mere featherweight," he assured him. "And is your

father nearly well again, Anne? He has written to me, but he did not
say much about himself."
"Papa looks much better than you do, M. le Chevalier," said the
little boy critically. "He can walk quite well now. He is coming to see
you when he is quite better. Grandpapa is downstairs, you know; he
will come up soon, I expect."
La Vireville, in his turn, surveyed the visitor perched on his body.

Anne's legs, in their blue pantaloons, stuck out straight in front of
him on the bed; the shoes at the end of them looked ridiculously
small. His curls, falling on his deep ruffle, seemed heavier and a little
longer than of yore, and the sun was busily employed in gilding
them. For the first time, therefore, La Vireville was really conscious
of the presence of that luminary.
Anne-Hilarion was the first to break the silence. "Did Papa tell
you in his letter," he inquired, "that a lady came into the garden to
ask for you, M. le Chevalier?"
"A lady!" exclaimed Fortuné. "What garden, child—here?"
"Oh no," replied Anne. "The garden in the Square at home. It is a

long time ago now. I was there, and John Simms, and I had leaves
in my cart—dead leaves—and she came in, all in black, and she
asked if you had come back from France, and I said no, and then I
cried, and I think she cried too, and she kissed me, and then
Grandpapa came, and——"
"Stop a moment!" cried La Vireville, who was not without
experience of the volume of detail Anne could pour forth when once
he was embarked on the tide of narration. "What was the lady like?
Was she young?"
"She was not so old as Elspeth," pronounced the Comte de

Flavigny, after due thought.
La Vireville gave a rather shaken laugh. Had this impossible thing

really happened? Anne-Hilarion never lied. But—it must have been
someone else!
"You did not learn her name, I suppose, child?" he asked, his
heart beginning to thud.
"Yes," responded Anne. "I asked Grandpapa afterwards, because

I liked her. She was called the Comtesse de Préfontaine—or perhaps
it was Guéfontaine."
La Vireville's heart missed a couple of beats, then pounded

harder than ever, seeming to shake his whole body—a humiliating
experience. But for his present physical condition, however, no doubt
he would not have gasped for breath as he did, nor would the colour
have come and gone like a woman's in his hollow cheeks.
Nevertheless, as both these things happened, Anne-Hilarion looked
at him in a little dismay.
"You are—do you feel ill, M. le Chevalier?" he asked solicitously.
"No—yes," stammered Fortuné, lifting himself on his elbow. "No,

child, don't move! It is not that you are . . . too heavy." He drew a
long breath, closed his eyes, and dropped back on his pillows. "What
did you and Grandpapa tell Mme. de Guéfontaine?" he asked, after a
moment.
"Grandpapa told her, I think, that the Republican soldiers had

shot you at that place—Quiberon. . . . M. le Chevalier," continued
Anne, leaning with very wide-open eyes towards him, and thus still
further contributing to the discomfort of the leg on which he was
situated, "did they really and truly shoot you?"
"Well," said Fortuné, in a dreamy voice, "they did and they did
not. Anyhow, here I am, you see." He reopened his eyes.
"Yes," said Anne, in a tone of great contentment, and bestowed

on his friend one of those infrequent smiles of his, sudden and shy,
at the same time sliding his hand into the strong, wasted one lying
idly near it on the coverlet.
A thrill ran up Fortuné's arm. Ever since he had seen Anne

standing by the door he had been conscious of a strange sensation,
as if, with the advent of the child who had sailed the seas with him
on that wild adventure, there had begun to blow through his hot
brain the first whisper of a clean and joyous breath that came from
the waves themselves. A moment or two ago, those lips had made
him an annunciation the full meaning of which he could hardly
grasp. Now, at the little warm touch, his eyes were suddenly
opened, and he saw the tainted memory in his heart as only a
handful of dead dust, not worth the keeping, fit only to be scattered
on that wind of morning. It was the past, useless and done with, a
thing long dead. . . . Here, close to him, touching him, smiling at
him, enshrined in this child who had no past, shone the future, like
something gallant and green with the dew on it, and, blowing over
it, that strong, fresh wind. The worthless burden he had carried for
years fell from him. He too could have what René de Flavigny had,
the air of morning at his gates—nay, morning's self. . . .
In the almost physical sense of deliverance he must have gripped

Anne-Hilarion's hand very hard, for the child gave a movement.
"Little pigeon, have I hurt you?" cried La Vireville, instantly

penitent, releasing the imprisoned hand. "I am so sorry. . . . I did
not know that I was still so strong." At the moment, indeed, he
hardly knew anything—scarcely, even, what he was saying.
Anne-Hilarion carried the injured member to his mouth for a

second or two, then he put it back in La Vireville's palm. "I am very
glad that you are so strong, M. le Chevalier," he said valiantly.
"Perhaps you will soon be quite well again. I hope so very much.
And then—what are you going to do?"
"What am I going to do, Anne?" A little while ago, under that

cloud of lassitude and depression, the question would have seemed
a mockery. "Well, you know—or you will soon know—that they had
to cut off my other arm, but I can still hold a sword—and hurt a
small boy's hand, eh? When I get quite better I shall go back to the
heather, and the sea, and perhaps . . ." He broke off and fell silent,
staring at his visitor with an air compound of bewilderment and
meditation. "Meanwhile, am I not to see the new goldfish?"
Anne-Hilarion slipped promptly from the bed and ran to the

corner by the door. Anon, raising himself from his stooping position,
and carrying it between his hands with even more than his
accustomed care, he came back with his trophy. His eyes were very
bright.
"It is my biggest one of all," he observed, as La Vireville propped

himself on his elbow to view the captive. "I called it after you, M. le
Chevalier . . . you do not mind? And I thought, as you were ill . . .
and I heard Papa and Grandpapa say you could never be repaid for
coming after me to France . . . you might . . . I mean I brought it to
give it to you, if you liked . . . for your own!"
"Oh, my child," said La Vireville, rather breathlessly, "you have

given me much more than that!"
BOOK FOUR
THE OLDEST ROAD OF ALL
"Be it wind, be it weet, be it hail, be it sleet,
Our ship must sail the faem;
The king's daughter o' Noroway,
'Tis we must fetch her hame."
Sir Patrick Spens.
"All quests end here, all voyagings, all ventures:

Is not my white breast haven to your sail?"
The Wave's Song.
CHAPTER XXXVIII
Flower of the Gorse
(1)
A brilliant May morning of sun and wind was exulting over the
beautiful harbour of St. Peter Port at Guernsey, and over the old
town rising steeply like an amphitheatre from its blue waters. But
the aged salt who was making his way up one of the narrow streets
with a basket of freshly caught lobsters on his arm was not
particularly responsive to the sunshine; indeed, the air with which he
paused and mopped his red face suggested that an injured "Very hot
for the time of year!" would issue from his bearded mouth in
response to any greeting.
As he put away his bandana and prepared to resume his ascent

of the cobbles, he observed two persons coming down, one behind
the other—a young man in uniform, and, in front of him, a girl in the
old Guernsey costume of chintz-patterned, quilted gown, opening in
front over the black stuff petticoat into the pocket-holes of which,
after the island fashion, it was tucked. This damsel came tripping
down, despite the steepness of the street, happy no doubt in the
conviction that the officer behind her was admiring her trim feet and
ankles in their blue stockings and buckled black velvet shoes.
Unfortunately the officer could not see her pretty face, framed in a
close mob cap under an ugly bonnet with enormous bows. Only the
ascending fisherman, at the moment, had a sight of that, and yet his
gaze was fixed precisely on the soldier behind her, scenting a
possible purchaser in him rather than in the native maiden. And the
officer, too, seemed to have his eye on the fisherman, and slackened
his pace as he came nearer. So Beauty, casting half a glance on the
writhing lobsters, passed unheeded.
"I suppose you can't tell me, mon vieux, the name of that vessel
just come into harbour?" asked the officer, stopping.
The uniform was English, but the wearer did not look quite
English, and he spoke in French. As a native of the Channel Islands
the ancient mariner accosted should have understood that tongue,
but for purposes of his own he affected not to do so.
"Very fine they are, indeed, sir!" he replied, peering into his
basket. "Comes from the rocks over by L'Etac, they do. You wants to
now the price? Well, this one——" and he held out a freckled ebony
form that slowly waved its spectral antennae at the young officer.
The latter pushed it aside with an impatient cane. "No, no—I

don't want one of those things to-day. I wish to know what that
vessel down there is—and I am sure you understand me perfectly!"
Having observed with one eye that the officer's other hand was
moving in the direction of his waistcoat pocket, the seafarer turned
both in the direction of the Frenchman's pointing cane. "Ah, yon,
just about to make fast," he said, pointing, too, with the rejected
lobster. "She'll likely be the Government sloop Cormorant, bound for
Jersey, come in here with despatches. Thank you, sir! And you won't
take this beauty home to your good lady?"
But the young officer shook his head with a smile, and continued
his downward path to the harbour.
Although to many dwellers in a port, especially in an island port,

the mere arrival and identification of a vessel is in itself a matter of
interest, this young Frenchman had a particular reason for
questioning the fisherman. The major of his regiment was ill;
medicaments had been ordered from England, and Lieutenant Henri
du Coudrais, finding himself unoccupied after an early parade, had
offered, on news of the arrival of a sail, to go down to the harbour,
instead of the major's servant, and ascertain if the drugs in question
had come.
But the sloop, when he got down to the quayside, had only just
finished making fast. Evidently she had a passenger, for he observed
among the sailors on her deck a tall man in a grey redingote, whose
general appearance seemed, somehow, to be familiar. But he could
not see his face, and thought no more about him till in a few
moments he came over the gang-plank.
And then, in one and the same instant, Henri du Coudrais saw
that the passenger's left sleeve was pinned to his breast, and
recognised him. A second later, and he had himself been recognised
by those keen eyes.
"M. du Coudrais!" called out the newcomer. "What good fortune

brings you here? I was just about to ask my way to your lodging."
The young officer had been stricken dumb for a moment. "M. de
la Vireville!" he exclaimed at last. "Is it possible?"
"Why not?" asked La Vireville, holding out his hand. "But I

suppose you thought that I was dead?"
"Indeed we did!" confessed his compatriot, grasping the

proffered hand warmly. "After many inquiries, we were convinced at
last. Then you escaped after all! But I am sorry to see . . ."
"Oh, I left that at Houat," said Fortuné composedly. "An

unnecessary luxury, two arms, I assure you, mon ami. I cannot think
how I ever found work for both. You are surprised to see me? Well, I
am on my way to Jersey; the sloop sails again this afternoon. I came
by her because she was touching here, and I wished to wait upon
you and Mme. de Guéfontaine."
"For my part, Chevalier, I am delighted to see you," said du

Coudrais, with much cordiality, "and I hope you will do me the
honour of dining with me; but my sister, I am sorry to say, is not
with me for the moment. She is over in Sark."
"In Sark?" repeated Fortuné, surprised, looking instinctively over
the intense blue to where, six or seven miles away, the little island
floated like a rock-set jewel. "When will she return?"
"Not until to-morrow or the day after, I am afraid," answered her

brother. "She has gone over to see a poor émigré family settled in a
farm there—that of an old nurse, in fact. She generally spends a
night or two with them. I need not tell you, Chevalier, how sorry she
will be to miss you. Could you not stay here till her return? The
hospitality that I can offer you is not very sumptuous, but I should
be deeply honoured by your acceptance of it."
Fortuné bit his lip thoughtfully, still looking over the sea to Sark.
Then he shook his head. "I thank you a thousand times, but I
cannot stay. I am awaited at Jersey. . . . Will you give me a word in
private, M. du Coudrais?—over there, for instance, at the end of the
jetty, would serve."
"Willingly," said Raymonde's brother, and followed him.
"You may possibly guess, Monsieur," began La Vireville, still

preoccupied with the sight of Sark, "why I wish to wait upon Mme.
de Guéfontaine?"
The young officer took on a very discreet air. "You are, perhaps,
in need of an agent de la correspondance over there again?"
La Vireville smiled. "Of that—and more," he said. "God knows I

have little enough to offer her—probably she won't even look at me
—but I should be glad to know that I had your consent to address
your sister."
"M. le Chevalier," retorted Henri du Coudrais, "do you suppose

that I have forgotten last April? I have not met any man to whom I
would sooner commit my sister. As for Raymonde—but she must
speak for herself."
"You are very kind, du Coudrais," said La Vireville, but he sighed.
"I wish I could think your sister would be as easily pleased. . . . It is
only right to point out to you that I have neither money, nor
prospects, nor a home, nor even two arms, to offer her——"
"But you asserted just now that one was sufficient," observed
Henri du Coudrais, leaning back with a smile against the rail that ran
out to the beacon light. "As for fortune or prospects, which of us
émigrés has those nowadays? And upon my soul, I don't know a
woman on earth who is less set on either than Raymonde."
"I suppose that I ought not to ask if there is any other man?"
"There was the Duc de Pontferrand; she refused him last October
—just at the time, Monsieur Augustin, when she was making
inquiries about you in London from the old gentleman whose name I
cannot remember, who lives with a little boy in Cavendish Square."
"I know she did that, God bless her!" said La Vireville. "I did not,
of course, know about the Duc." He fell silent, fingering the rail and
still gazing out to sea. It occurred to du Coudrais that though he had
the look of one who has weathered a long and trying illness, he yet
seemed in some indefinable way a younger man.
"Why should I not hire a boat and sail over to Sark?" asked La
Vireville suddenly. "My wooing must in any case be rough and ready.
I could be back before the Cormorant sails, if I went at once."
"Ma foi, an excellent idea!" said Raymonde's brother heartily.

"That is, of course, the solution. I will procure you a boat, if you
wish. You must be sure to take a native with you, even though the
distance be not great, for sailing hereabouts is dangerous, if only on
account of the hidden rocks—'stones,' as they call them." He looked
about him. "There is Tom Le Pelley; he would serve your purpose."
(2)
A quarter of an hour later La Vireville was sailing over that
laughing expanse towards the gem of rose and emerald and flame,
whose beauty, though his eyes were set upon it all the while, he
hardly marked. The boatman spoke of channels and swift tides, of
the Anfroques, the Longue Pierre, the Goubinière, but names of
reefs and rocks went by La Vireville unheeded. He was going to put
to the test what Anne-Hilarion had shown him. He was liberated at
last from his servitude of mind, and he wanted Raymonde—wanted
her with all his heart. It was very strange to him now that he had
not known this when he was with her more than a year ago.
Du Coudrais had given him the name of the farm which Mme. de
Guéfontaine had gone to visit, and once landed he found it easily
enough, for there were not many of them on that slender strip of an
isle, pillared on its rocks and magic caves. But Raymonde was not
there, and they told him that she was out on one of the headlands.
And there, after a space, he found her, among the golden brands
of the gorse, looking out to sea in the direction of the coast of
France. The wind blew against her; she shaded her eyes with her
hand under her little three-cornered hat, as from the lovely land of
exile she gazed intently at a dearer shore. She did not see him, nor,
from the talk of the wind in her ears, hear his footsteps brushing
through the gorse—and Fortuné stopped short, for now that he
beheld her again with his bodily eyes he knew that his desire for her
was even greater than he had thought, and in proportion the fear
swelled in him to conviction that so great a gift could never be
meant for him. So he stood there bareheaded in the sunshine, his
heart mingled flame and water, aching to see her hidden face, and
yet afraid to put his destiny to the touch. But at last, since she was
still unconscious of his presence, he was forced to make it known.
"Madame!"
And at that she turned round with a start. Colour swept over her
face and was gone again, and in her eyes there was something that
was almost fear.
"Monsieur . . . de la Vireville!" she exclaimed, on a sharp catch of

the breath.
It was the first time, as he instantly realised, that she had ever
called him by his name, that name which was dipped for her in such
painful memories.
"Me voici!" said he, and casting his hat on to a gorse bush
advanced to kiss her hand.
"I . . . I am not sure . . . that you are not a ghost!" she said, not
very steadily, as she surrendered it.
"Indeed I am not!" he unnecessarily assured her, for the kiss he

put on it must have convinced her that he was flesh and blood, and
perhaps the wave of colour which once more dyed her face derived
its temperature from the warmth of that salutation.
"But you . . . M. le Chevalier, but you have returned from the

dead! They told me you had been shot!"
"Yes, I have returned from the dead," agreed Fortuné—"for a

purpose."
She did not ask what the purpose was; she still seemed shaken,
uncertain of herself and of him. But her gaze, swift and
compassionate, swept over everything that the sunlight showed so
relentlessly—the traces of past suffering on his face, the added grey
at his temples, and the pinned-up sleeve.
"Ah, que vous avez dû souffrir!" she said to herself. Then she put
her hand to her head, as if she still felt herself in a dream.
"But where have you come from, M. le Chevalier?" she asked.

"And why are you here, in Sark?"
He looked at her full, and answered bluntly, "To ask you to marry
me!"
But as, giving an exclamation, she turned away, he hastily

abandoned this ground.
"I have nothing to offer you, Madame," he went on quickly.

"Neither money nor position nor a home, nor even two arms to
defend you. The Republic has taken all those. And—for I am
determined to be very frank with you—I must tell you that for ten
years my heart has not been my own to offer. It was pledged to a
memory. It has come back to me now, thank God, but I fear it has
the dust of those years on it, and I am no longer very young." He
paused a moment, and the sea of Sark, that is for ever booming in
its enchanted caverns, gave a dull echo to his words. "It is because
you too, Raymonde, have greatly loved and hated—I happen, do I
not, to know how much?" he added, with the shadow of a smile
—"that I am thus open with you. But my old love and hate are both
over and done with now. I have a new, a better love—and it is all
yours, as long as I shall live."
Mme. de Guéfontaine was examining a single childish bloom of

gorse, just outgrowing its rough yellow-brown pinafore. And she said
nothing.
"I have no time to wrap this meagre offer in fair phrases," went
on Fortuné. "I doubt if they would improve it, and you are not, I
think, the woman to care for them. I can only say this over and over
again, that I love you and that I want you. It was you—the thought
of you—that saved me at Quiberon; I used to dream of you at
Houat. Raymonde!"
Still she did not answer, and stood with her head averted.
"Raymonde," said he, coming a little nearer, mingled command

and entreaty in his tone, "for God's sake put down that flower and
answer me!—only do not send me back to France with a refusal! If
you cannot make up your mind to-day—and I must crave your
forgiveness indeed for so blunt and hasty a wooing—at least let me
take back with me a glimmer of hope!"
At that she looked up. Her face was transfigured, but he dared
not try to interpret its new meaning.
"You are going back to France, in spite of everything, to that old

life of peril and hardships?"
"Of course," said he. "But if you would accept it, I should have a
home to offer you in Jersey. And when better days come——"
She interrupted him. "You misunderstand me, M. le Chevalier. I

should not marry any man who was risking his life over there, to
stay behind myself in safety. A wife's place, if she can help him, is
with her husband." A smile wove itself into the beautiful radiance.
"Shall you not need an agent at Kerdronan?"
For a second the gorse heaved beneath him. "Do you mean what
you are saying?" cried La Vireville, seizing her wrist. "Will you really
marry a penniless cripple who has nothing but his sword?"
Her smile was brilliant now, and dazzled Fortuné while she faced
him, captive, as on a certain morning a year ago. "No, M. de la
Vireville, I shall marry a man! As you know, for three years I had
hated your name. But, as you wear it, I have long seen that I could
not take a nobler."
So the woman he desired lay at last against Fortuné de la

Vireville's breast, and up from the sea of gorse in which they stood
welled the warm honey-sweet scent that is like no other in the world
to steal away the heart. The wind had dropped to a caress; it caught
at Raymonde's gown no longer, and out over the illimitable wrinkled
blue, from the height on which they stood, the poised gulls looked
like slowly drifting flecks of snow.
But out over there was also the long purple line of Jersey, and his
pledged word. Time was all too short. As long as he lived the scent
of gorse would always bring this hour to him, but the actual hour
itself was measured with very few sands.
"Will you come back with me now to Guernsey, to your brother,

Raymonde?" he asked softly, stooping his head.
"Yes," she answered, without moving. Her voice sounded like a

voice in a dream.
"And I will return from Jersey, and we will be married at once?"
"Yes," she said again.
"My God, I can't believe it!" said Fortuné to himself, and kissed
her once more.
So they went together to the little farm, itself named from the
gorse, the Clos-ajonc, to tell her pensioners that she was leaving
them immediately. And, no doubt to show that she did not consider
him so maimed as to be incapable of affording her support,
Raymonde leant all the way upon his arm.
CHAPTER XXXIX
Flower of the Foam
La Vireville did not go into the little Clos-ajonc with his lady. He
waited for her outside, leaning upon its low, whitewashed wall, over
which the tamarisk whispered with its feathery foliage. Breaths of
the gorse came to him even here, and the whole warm air was
vibrating with the lark's ecstasy. And Fortuné could hardly believe his
happiness, so strange a thing to him. Old dreams, long put away,
came back to him, merged in the new. Had he not yearned
sometimes, despite himself, to have, in what remained of this hard
and shifting existence of his, brief enough in its pleasures but
endless in its unceasing fatigue and peril and anxiety—the life that
was often no better than a hunted animal's—to have one place that
was home, and shrine, and star? Well, he had his desire now; he
had won that place, that heart, at last.
Yet even as he leant there, absorbed in contemplation, his mind

was suddenly pierced with a most evil arrow of a thought. What if
Raymonde had taken him out of pity, as a woman sometimes will . .
. or worse, out of a sense of gratitude?
The idea assailed him so unexpectedly, and so much from

without his own consciousness, that La Vireville dropped the strand
of tamarisk which he was idly fingering, and started up,
straightening himself as under an actual physical blow. Good God, it
was impossible! "No," said a tiny derisive voice in his heart, "far from
it! It is very possible. Even as to you yourself she is but a makeshift
—you told her so with your own lips—so to her you are only a man
for whom she is sorry, and to whom she is under an obligation. So
much the better for you! And what else did you expect?"
And part of this two-sided onslaught La Vireville instantly and

furiously repelled. No, no, it was a lie—she was not a makeshift! If

Manias Panics and Crashes A History of Financial Crises 7th download pdf

Uploaded by

Copyright:

Available Formats

Manias Panics and Crashes A History of Financial Crises 7th download pdf

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Manias Panics and Crashes A History of Financial Crises 7th download pdf

Uploaded by

Copyright:

Available Formats

Download the full version of the ebook at

An Introduction to Applied Multivariate

Explore and download more ebook at https://ebookultra.com

An Introduction to Multivariate Statistical Analysis Wiley

Introduction to Applied Statistical Signal Analysis 3rd

An introduction to applied linguistics Third Edition.

An Introduction to Harmonic Analysis 3rd Edition Yitzhak

An Introduction to Statistical Genetic Data Analysis

Analysis of Multivariate Social Science Data Second

Introduction to Stochastic Calculus Applied to Finance

Cluster Analysis and Data Mining An Introduction King

Tenko Raykov George A. Marcoulides

New York London

© 2008 by Taylor & Francis Group, LLC

Printed in the United States of America on acid-free paper

International Standard Book Number-13: 978-0-8058-6375-8 (Hardcover)

Library of Congress Cataloging-in-Publication Data

Introduction to applied multivariate analysis / by Tenko Raykov & George A.

Visit the Taylor & Francis Web site at

and the Psychology Press Web site at

Chapter 1 Introduction to Multivariate Statistics

Chapter 2 Elements of Matrix Theory

Chapter 3 Data Screening and Preliminary Analyses

Chapter 4 Multivariate Analysis of Group Differences

Chapter 5 Repeated Measure Analysis of Variance

Chapter 6 Analysis of Covariance

Chapter 7 Principal Component Analysis

7.5.1 Multicollinearity Detection ..................................................... 234

Chapter 8 Exploratory Factor Analysis

Chapter 9 Conﬁrmatory Factor Analysis

Chapter 10 Discriminant Function Analysis

Chapter 11 Canonical Correlation Analysis

Chapter 12 An Introduction to the Analysis

Chapter 13 Multivariate Analysis of Change Processes

Author Index ..................................................................................... 473

Subject Index ..................................................................................... 477

Having taught applied multivariate statistics for a number of years, we

One of the simplest conceptual deﬁnitions of multivariate statistics (MVS)

1.1 Deﬁnition of Multivariate Statistics

2 Introduction to Applied Multivariate Analysis

populations. Consequently, in empirical behavioral and social studies, one

Introduction to Multivariate Statistics 3

question appears to be relatively simple and in fact may be addressed

4 Introduction to Applied Multivariate Analysis

By way of contrast, MVS may be viewed as an extension of UVS in the

Introduction to Multivariate Statistics 5

devoted to their discussion. The common theme unifying the research

1.2 Relationship of Multivariate Statistics

6 Introduction to Applied Multivariate Analysis

Introduction to Multivariate Statistics 7

1.3 Choice of Variables and Multivariate Method,

8 Introduction to Applied Multivariate Analysis

affected by whether the data stem from an experimental or an observational

and has special optimality properties (Y* is occasionally referred to as

1.4 Data for Multivariate Analyses

Introduction to Multivariate Statistics 9

A data matrix is a rectangular array of collected (recorded) scores from

10 Introduction to Applied Multivariate Analysis

subjects. This assumption may be violated in cases where there is a

Introduction to Multivariate Statistics 11

GPA Pearson Correlation 1 .387* .659 .453

IQ Pearson Correlation .659 .206 1 .784

HOURS_TV Pearson Correlation .453 .125 .784 1

GPA Pearson Correlation 1 .387* .040 .659 .453