Download the full version of the ebook at

https://ebookfinal.com

Using Multivariate Statistics Barbara G.

Tabachnick

https://ebookfinal.com/download/using-
multivariate-statistics-barbara-g-tabachnick/

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

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Exploratory Multivariate Analysis by Example Using R

Second Edition Husson

https://ebookfinal.com/download/exploratory-multivariate-analysis-by-
example-using-r-second-edition-husson/

ebookfinal.com

Multivariate Statistics High Dimensional and Large Sample

Approximations 1st Edition Yasunori Fujikoshi

https://ebookfinal.com/download/multivariate-statistics-high-
dimensional-and-large-sample-approximations-1st-edition-yasunori-
fujikoshi/
ebookfinal.com

Applied Statistics From Bivariate Through Multivariate

Techniques 2nd Edition Edition Rebecca M. Warner

https://ebookfinal.com/download/applied-statistics-from-bivariate-
through-multivariate-techniques-2nd-edition-edition-rebecca-m-warner/

ebookfinal.com

Discovering Statistics using SPSS 4th Edition Andy Field

https://ebookfinal.com/download/discovering-statistics-using-spss-4th-
edition-andy-field/

ebookfinal.com
Discovering Statistics Using SPSS 3rd Edition Andy Field

https://ebookfinal.com/download/discovering-statistics-using-spss-3rd-
edition-andy-field/

ebookfinal.com

Teaching Statistics Using Baseball 2nd Edition James

Albert

https://ebookfinal.com/download/teaching-statistics-using-
baseball-2nd-edition-james-albert/

ebookfinal.com

Using Biographical Methods in Social Research 1st Edition

Barbara Merrill

https://ebookfinal.com/download/using-biographical-methods-in-social-
research-1st-edition-barbara-merrill/

ebookfinal.com

Introductory Statistics for Health and Nursing Using SPSS

Louise Marston

https://ebookfinal.com/download/introductory-statistics-for-health-
and-nursing-using-spss-louise-marston/

ebookfinal.com

Teaching Music Through Composition A Curriculum Using

Technology 1st Edition Barbara Freedman

https://ebookfinal.com/download/teaching-music-through-composition-a-
curriculum-using-technology-1st-edition-barbara-freedman/

ebookfinal.com
Using Multivariate Statistics Barbara G. Tabachnick
Digital Instant Download
Author(s): Barbara G. Tabachnick, Linda S. Fidell
ISBN(s): 9780205459384, 0205459382
Edition: None
File Details: PDF, 40.85 MB
Year: 2007
Language: english
 FIFTH EDITION

Using Multivariate Statistics

Barbara G. Tabachnick
California State University, Northridge

Linda S. Fidell
California State University, Northridge

Boston rn New York rn San Francisco

Mexico City Montreal rn Toronto rn London rn Madrid rn Munich Paris
Hong Kong rn Singapore rn Tokyo rn CapeTown rn Sydney
Executive Editor: Siisczrz Hlzrttnan
Editorial Assistant< Tizerese Felser
Marketing Manager: Karen Nat~lle
Senior Production Administrator: Donna Simons
Editorial Production Service: Omegatype T y p o g m p h ~Inc.
Composition Buyer: Andrew Erryo
Manufacturing Buyer: Andrew T~lrso
Electronic Composition: Omegcztype Typography, Inc.
Cover Administrator: Joel Geizdron

For related titles and support materials, visit our online catalog at www.ablongman.com.

Copyright 0 2007,2001, I996 Pearson Education. Inc.

All rights reserved. No part of the material protected by t h ~ scopyright notice may be reproduced
or utilized in any form or by any means. electronic or mechanical, including photocopying, recording,
or by any information storage and retrieval system, without written permission from the copyright holder.

To obtain permissionis) to use materials from this work, please submit a written request to Allyn and Bacon,
Permissions Department, 75 Arlington Street, Boston, MA 02 116, or fax your request to 6 17-848-7320.

Between the time web site information is gathered and published. it is not unusual for some sites to have
closed. Also, the transcription of URLs can result in typographical errors. The publisher would appreciate
notification where these occur so that they may be corrected in subsequent editions.

ISBN 0-205-45938-2

Punted In the Un~redState\ o t Amerlca

1 0 9 8 7 6 5 4 3 2 1 RRD 10 09 08 07 06
CONTENTS

Preface xxvii

1.1 Multivariate Statistics: Why? 1

1.1.1 The Domain of Multivariate Statistics: Numbers of IVs
and DVs 1
1.1.2 Experimental and Nonexperimental Research 2
1.1.3 Computers and Multivariate Statistics 4
1.1.4 Garbage In, Roses Out? 5
1.2 Some Useful Definitions 5
1.2.1 Continuous. Discrete, and Dichotomous Data 5
1.2.2 Samples and Populations 7
1.2.3 Descriptive and Inferential Statistics 7
1.2.4 Orthogonality: Standard and Sequential Analyses 8
1.3 Linear Combinations of Variables 10
1.4 Number and Nature of Variables to Include 11
1.5 Statistical Power 11
1.6 Data Appropriate for Multivariate Statistics 12
1.6.1 The Data Matrix 12
1.6.2 The Correlation Matrix 13
1.6.3 The Variance-Covariance Matrix 14
1.6.4 The SUG-of-Sc;uaresand Cr~ss-ProductsMatrix l4
1.6.5 Residuals 16
1.7 Organization of the Book 16

3
li A Guide to Statistical Techniques: Using the Book 17
2.1 Research Questions and Associated Techniques 17
2.1.1 Degree of Relationship among Variables 17
2.1.1.1 Bivariate r 17
2.1.1.2 Multiple R 18
2.1.1.3 Sequential R 18
2.1.1.4 Canonical R I8
2.1.1.5 Multiway Frequency Analysis 19
7.1.1.6 Multilevel Modeling 19

iii
iv CONTENTS

2.1.2 Significance ot'(;i-oup Difference\ I9

3. I .Z. i One-Way XKOL'A and r Test 19
2.1.1.2 One-Way ANCOVA 30
2.1.2.3 Factorial ANOVA 20
2.1.2.4 Factorial ANCOVA 20
2.1.2.5 Hotelling's T' 21
2.1.2.6 One-Way MANOVA 21
2.1.2.7 One-Way MANCOVA 21
2.1.2.8 Factorial MANOVA 22
2.1.2.9 Factorial MANCOVA 22
2.1.2.10 Profile Analysis of Repeated Measures 23
2.1.3 Prediction of Group Membership 23
2.1.3.1 One-Way Discriminant 23
2.1.3.2 Sequential One-Way Discriminant 24
2.1.3.3 Multiway Frequency Analysis (Logit) 24
2.1.3.4 Logistic Regression 24
2. I .3.5 Sequential Logistic Regression 25
2.1.3.6 Factorial Discriminant Analysis 25
2.1.3.7 Sequential Factorial Discriminant Analysis 25
2.1.4 Structure 25
2.1.4.1 Principal Components 25
2.1.3.2 Factor Analysis 26
2. !.4.? Structural Eqtiation Modeling 26
2.1.5 Time Course of Events 26
2.1.5.1 Survival/Failure Analysis '26
2.1.5.2 Time-Series Analysis 27
2.2 Some Further Comparisons 27
2.3 A Decision Tree 28
2.4 Technique Chapters 31
2.5 Preliiiiinary Check of the Data 32

3 Review of Univariate and Bivariate Statistics 33

3.1 Hypothesis Testing 33
3.1.1 One-Sample z Test as Prototype 33
3.1.2 Power 36
3.1.3 Extensions of the Model 37
3.1.4 Controversy Surrounding Significance Testing 37
3.2 Analysis of Variance 37
3.2.1 One-Way Between-Subjects ANOVA 39
3.2.2 Factorial Between-Subjects ANOVA 42
3.2.3 Within-Subjects ANOVA 43
3.2.4 Mixed Between-Within-Subjects ANOVA 46
 3.2.5 De\ign Complexity 47
3.2.5.1 Nesting 47
3.2.5.1 Latin-Square Designs 47
3.2.5.3 Unequal 11 and Nonorthogonal~ty 48
3.2.5.3 Fixed and Random Effects 49
3.2.6 Specific Comparisons 49
3.2.6.1 Weighting Coefficients for Comparisons 50
3.2.6.2 Orthogonality of Weighting Coefficients 50
3.2.6.3 Obtained F for Comparisons 51
3.2.6.4 Critical F for Planned Comparisons 52
3.2.6.5 Critical F for Post Hoc Comparisons 53
3.3 Parameter Estimation 53
3.4 Effect Size 54
3.5 Bivariate Statistics: Correlation and Regression 56
3.5.1 Correlation 56
3.5.2 Regression 57
3.6 Chi-Square Analysis 58

Cleaning Up Your Act: Screening Data

Prior to Analysis 60
4.1 Important Issues in Data Screening 61
4.1 . 1Accuracy of Data File 61
4.1.2 Honest Correlations 61
4.1.2.1 Inflated Correlation 61
4.1.2.2 Deflated Correlation 61
4.1.3 Missing Data 62
4. !.3. ! Deleting Cases or Variables 63
4.1.3.2 Estimating Missing Data 66
4.1.3.3 Using a Missing Data Correlation Matrix 70
4.1.3.4 Treating Missing Data as Data 71
4.1.3.5 Repeating Analyses with and without Missing Data 71
4.1.3.6 Choosing among Methods for Dealing
with Missing Data 71
4.1.4 Outliers 72
4.1.4.1 Detecting Univariate and Multivariate Outliers 73
4.1.4.2 Describing Outliers 76
4.1.4.3 Reducing the Influence of Outliers 77
4.1.4.4 Outliers in a Solution 77
4.1.5 Normality, Linearity, and Hornoscedasticity 78
4.1.5.1 Normality 79
4.1.5.2 Linearity 83
4.1.5.3 Homoscedasticity, Homogeneity of Variance. and
Homogeneity of Variance-Covariance Matrices 85
vi CONTENTS

4.1 (7 Cornnion Data Tr;tn(forlndti~n\ 86

'ind Sing~11~1-1ty 88
1.1.7 M~~lticollinedrit>
4.1.8 A Checklist and Some Pract~calRecommendation\ 9I
4.2 Complete Examples of Data Screening 92
4.2.1 Screening Ungrouped Data 92
4.2.1.1 Accuracy of Input. Missing Data, Distributions,
and Univariate Outliers 93
4.2.1.2 Linearity and Homoscedasticity 96
4.2.1.3 Transformation 98
4.2.1.4 Detecting Multivariate Outliers 99
4.2.1.5 Variables Causing Cases to Be Outliers 100
4.2.1.6 Multicollinearity 104
4.2.2 Screening Grouped Data 105
4.2.2.1 Accuracy of Input, Missing Data, Distributions,
Homogeneity of Variance, and Univariate Outliers 105
4.2.2.2 Linearity 110
4.2.2.3 Multivariate Outliers 11 1
4.2.2.4 Variables Causing Cases to Be Outliers 1 13
4.2.2.5 Multicollinearity 1 14

Ir
3 Multiple Regression 117
5.1 General Purpose and Description 117
5.2 Kinds of Research Questions 118
5.2.1 Degree of Relationship 1 19
5.2.2 Irr~portanceof i'v'a 119
5.2.3 Adding IVs 119
5.2.4 Changing 1Vs 120
5.2.5 Contingencies among IVs 120
5.2.6 Comparing Sets of IVs 120
5.2.7 Predicting DV Scores for Members of a New Sample 120
5.2.8 Parameter Estimates 121
5.3 Limitations to Regression Analyses 121
5.3.1 Theoretical Issues 122
5.3.2 Practical Issues 123
5.3.2.1 Ratio of Cases to IVs 123
5.3.2.2 Absence of Outliers among the IVs and on the DV 124
5.3.2.3 Absence of Multicollinearity and Singularity 124
5.3.2.4 Normality, Linearity, Homoscedasticity of Residuals 125
5.3.2.5 Independence of Errors 128
5.3.2.6 Absence of Outliers in the Solution 128
5.4 Fundamental Equations for Multiple Regression 128
5.4.1 General Linear Equations 129
5.4.2 Matrix Equations 13 1
5.4.3 Computer Analyses of Small-Sample Example 134
 C O \ \ , E L r‘j vii

5.5 Major Types of klultiple Regression 136

5.5. l Stancl~rctMultiple Regression 136
5.5.2 Sequential Multiple Regression 138
5.5.3 Statistical (Stepwise) Regression 138
5.5.4 Choosing among Regression Strategies 143
. 5.6 Some Important Issues 144
5.6.1 Importance of IVs 144
5.6.1.l Standard Multiple Regression 146
5.6.1.2 Sequential or Statistical Regression 146
5.6.2 Statistical Inference 146
5.6.2.1 Test for Multiple R 147
5.6.2.2 Test of Regression Components 148
5.6.2.3 Test of Added Subset of IVs 149
5.6.2.4 Confidence Limits around B and Multiple" R 50
5.6.2.5 Comparing Two Sets of Predictors 152
5.6.3 Adjustment of R~ 153
5.6.4 Suppressor Variables 154
5.6.5 Regression Approach to ANOVA 155
5.6.6 Centering when Interactions and Powers of IVs
Are Included 157
5.6.7 Mediation in Causal Sequences 159
5.7 Complete Examples of Regression Analysis 161
5.7,1 Evaluation of Assumptions 161
5.7.1.1 Ratio of Cases to IVs 16 1
5.7.1.2 Normality, Linearity, Homoscedasticity,
and Independence of Residuals 16 1
5.7.1.3 Outliers 165
5.7.1.4 Multicollinearity and Singularity 167
5.7.2 Standard Multiple Regression 167
5.7.3 Sequential Regression 174
5.7.4 Example of Standard Multiple Regression with Missing Values
Multiply Imputed 179
5.8 Comparison of Programs 188
5.8.1 SPSS Package 188
5.8.2 SAS System 191
5.8.3 SYSTAT System 194

6 Analysis of Covariance 195

6.1 General Purpose and Description 195
6.2 Kinds of Research Questions 198
6.2.1 Main Effects of IV4 I98
6.2.2 Interactions among IVs 198
6.2.3 Specific Comparisons and Trend Analysis 199
6.2.4 Effects of Covariates 199
CONTENTS

6.2.5 Effect S ~ r e I99

6.2.6 Parameter E\t~rnates 199
6.3 Limitations to Analysis of Covariance 200
6.3.1 Theoretical Issues 200
6.3.2 Practical Issues 20 1
6.3.2.1 Unequal Sample Sizes, Missing Data, and Ratio of Cases
to IVs 201
6.3.2.2 Absence of Outliers 201
6.3.2.3 Absence of Multicollinearity and Singularity 20 1
6.3.2.4 Normality of Sampling Distributions 202
6.3.2.5 Homogeneity of Variance 202
6.3.2.6 Linearity 202
6.3.2.7 Homogeneity of Regression 202
6.3.2.8 Reliability of Covariates 203
6.4 Fundamental Equations for Analysis of Covariance 203
6.4.1 Sums of Squares and Cross Products 204
6.4.2 Significance Test and Effect Size 208
6.4.3 Computer Analyses of Small-Sample Example 209
6.5 Some Important Issues 211
6.5.1 Choosing Covariates 2 11
6.5.2 Evaluation of Covariates 2 12
6.5.3 Test for Homogeneity of Regression 2 13
6.5.4 Design Complexity 2 13
6.5.4.1 Within-Subjects and Mixed
Within-Between Designs 2 13
6.5.4.2 Unequal Sample Si7es 2 17
6.5.4.3 Specific Comparisons and Trend Analysis 2 18
6.5.4.4 Effect Size 22 1
6.5.5 Alternatives to ANCOVA 22 1
6.6 Complete Example of Analysis of Covariance 223
6.6.1 Evaluation of Assumptions 223
6.6. I. 1 Unequal n and Missing Data 224
6.6.1.2 Normality 224
6.6.1.3 Linearity 224
6.6.1.4 Outliers 224
6.6.1.5 Multicollinearity and Singularity 227
6.6.1.6 Homogeneity of Variance 228
6.6.1.7 Homogeneity of Regression 230
6.6.1.8 Reliability of Covariates 230
6.6.2 Analysis of Covariance 230
6.6.2.1 Main Analysis 230
6.6.2.2 Evaluation of Covariates 235
6.6.2.3 Homogeneity of Regression Run 237
6.7 Comparison of Programs 240
6.7.1 SPSSPackage 240
 CONTENTS ix

6 7 2 SXS Sqstem 710

6.7.3 SYSTAT Systern 310

7 Multivariate Analysis of Variance and Covariance 243

7.1 General Purpose and Description 243
7.2 Kinds of Research Questions 247
7.2.1 Main Effects of IVs 247
7.2.2 Interactions among IVs 247
7.2.3 Importance of DVs 247
7.2.4 Parameter Estimates 248
7.2.5 Specific Comparisons and Trend Analysis 248
7.2.6 Effect Size 248
7.2.7 Effects of Covariates 248
7.2.8 Repeated-Measures Analysis of Variance 249
7.3 Limitations to Multivariate Analysis of Variance
and Covariance 249
7.3.1 Theoretical Issues 249
7.3.2 Practical Issues 250
7.3.2.1 Unequal Sample Sizes. Missing Data, and Power 250
7.3.2.2 Multivariate Normality 25 :
7.3.2.3 Absence of Outliers 25 1
7.3.2.4 Homogeneity of Variance-Covariance Matrices 25 1
7.3.2.5 Linearity 252
7.3.2.6 Homogeneity of Regression 252
7.3.2.7 Reliability of Covariates 253
7.3.2.8 Absence of Multicollinearity and Singular~ty 253
7.4 Fundamental Equations for Multivariate Analysis
of Variance and Covariance 253
7.4.1 Multivariate Analysis of Variance 253
7.4.2 Computer Analyses of Small-Sample Example 261
7.4.3 Multivariate Analysis of Covariance 264
7.5 Some Important Issues 268
7.5.1 MANOVA vs. ANOVAs 268
7.5.2 Criteria for Statistical Inference 269
7.5.3 Assessing DVs 270
7.5.3.1 Univariate F 270
7.5.3.2 Roy-Bargmann Stepdown Analysis 27 1
7.5.3.3 Using Discriminant Analysis 272
7.5.3.4 Choosing among Strategies for Assessing DVs 273
7.5.4 Specific Comparisons and Trend Analysis 273
7.5.5 Design Complexity 274
7.5.5,l Within-Subjects and Between-Within Designs 273
7.5.5.3- Unequal Sample Sizes 276
X CONTENTS

7.6 Complete Examples of 3Iultivariate .Analysis of Variance

and Covariance 277
7.6.1 Evaluation of Assulnptions 177
7.6. I . I Unequal Sample Sizes and Missing Data 277
7.6. 1.2 Multivariate Normality 279
7.6.1.3 Linearity 279
7.6.1.4 Outliers 279
7.6.1.5 Homogencity of Variance-Covariance Matrices 280
7.6.1.6 Homogeneity of Regression 28 1
7.6.1.7 Reliability of Covariates 284
7.6.1.8 Multicollinearity and Singularity 285
7.6.2 Multivariate Analysis of Variance 285
7.6.3 Multivariate Analysis of Covariance 296
7.6.3.1 Assessing Covariates 296
7.6.3.2 Assessing DVs 296
7.7 Comparison of Programs 307
7.7.1 SPSS Package 307
7.7.2 SAS System 310
7.7.3 SYSTAT System 3 10

8 Profile Analysis: The Multivariate Approach

to Repeated Measures 311
8.1 General Purpose and Description 31 1
8.2 Kinds of Research Questions 312
8.2. I Parallelism of Profiles 3 12
8.2.2 Overali D~t'ferenceamong Groups 3 13
8.2.3 Flatness of Profiles 3 13
8.2.4 Contrasts Following Profile Analysis 3 13
8.2.5 Parameter Estimates 3 !3
8.2.6 Effect Size 314
8.3 Limitations to Profile Analysis 314
8.3.1 Theoretical Issues 3 14
8.3.2 Practical Issues 3 15
8.3.2.1 Sample Size, Missing Data, and Power 3 15
8.3.2.2 Multivariate Normality 3 15
8.3.2.3 Absence of Outliers 3 15
8.3.2.4 Homogeneity of Variance-Covariance Matrices 3 15
8.3.2.5 Linearity 3 16
8.3.2.6 Absence of Multicollinearity and Singularity 3 16
8.4 Fundamental Equations for Profile Analysis 316
8.4.1 Differences in Levels 3 16
8.4.2 Parallelism 3 18
 CONTENTS xi

8.4.7 Flatne\\ 32 I
8.4.1 Co~npi~ter
.Alialy\ss cif Small-S;\mplt: Example 323
8.5 Some Important Issues 329
8.5.1 Univariate vs. Multivariate Approach to Repeated Measures 329
8.5.2 Contrasts in Profile Analysis 33 1
8.5.2.1 Parallelism and Flatness Significant. Levels Not Signiticant
(Simple-effectsAnalysis) 333
8.5.2.2 Parallelism and Levels Signiticant, Flatness Not Signiticant
(Simple-effectsAnalysis) 336
8.5.2.3 Parallelism, Levels, and Flatness Significant
(Interaction Contrasts) 339
8.5.2.4 Only Parallelism Signiticant 339
8.5.3 Doubly-Multivariate Designs 339
8.5.4 Classifying Profiles 345
8.5.5 Imputation of Missing Values 345
8.6 Complete Examples of Profile Analysis 346
8.6.1 Profile Analysis of Subscales of the WISC 346
8.6.I . 1 Evaluation of Assumptions 346
8.6.1.2 Profile Analysis 35 1
8.6.2 Doubly-Multivariate Analysis of Reaction Time 360
8.6.2.1 Evaluation of Assumptions 360
8.6.2.2 Doubly-Multivzriate .4nalysis of Slope and Intercept 363
8.7 Comparison of Programs 37 1
8.7.1 SPSS Package 373
8.7.2 SAS System 373
8.7.3 SYSTAT System 37-4

9 Discriminant Analysis 375

9.1 General Purpose and Description 375
9.2 Kinds of Research Questions 378
9.2.1 Significance of Prediction 378
9.2.2 Number of Signiticant Discriminant Functions 378
9.2.3 Dimensions of Discrimination 379
9.2.4 Classification Functions 379
9.2.5 Adequacy of Classification 379
9.2.6 Effect Size 379
9.2.7 Importance of Predictor Variables 380
9.2.8 Significance of Prediction with Covariates 380
9.2.9 Estirtiation of Group Means 380
9.3 Limitations to Discriminant Analysis 381
9.3.1 Theoretical Issues 38 I
xii CONTENTS

9.7.2 Prilctic'al Ihsue\ 38 I

9.3.2. I Unequal Sample St/es. Missing Data. and Power 38 I
9.3.2.2 blultivat-iate Not-maltty 382
9.3.2.3 Absence of Outliers 382
9.3.1.4 Homogeneity of Variance-Covariance Matrices 382
9.3.1.5 Linearity 383
9.3.2.6 Absence of Multicollinearity and Singularity 383
9.4 Fundamental Equations for Discriminant Analysis 384
9.4.1 Derivation and Test of Discriminant Functions 384
9.4.2 Classification 387
9.4.3 Computer Analyses of Small-Sample Example 389
9.5 Types of Discriminant Function Analyses 395
9.5.1 Direct Discriminant Analysis 395
9.5.2 Sequential Discriminant Analysis 396
9.5.3 Stepwise (Statistical) Discriminant Analysis 396
9.6 Some Important Issues 397
9.6.1 Statistical Inference 397
9.6. I. I Criteria for Overall Statistical Significance 397
9.6.1.2 Stepping Methods 397
9.6.2 Number of Discriminant Functions 398
9.6.3 Interpreting Discriminant Functions 398
9.6.3.1 Discriminant Function Plots 398
9.6.3.2 Structure Matrix of Loadings 400
9.6.4 Evaluating Predictor Variables 401
9.6.5 Effectsize 402
9.6.6 Design Complexity: Factorial Designs 403
9.6.7 Use of Classification Procedures 404
9.0.7.! Cross-iralidatior; and Neiv Cases 405
9.6.7.2 Jackknifed Classification 405
9.6.7.3 Evaluating Improvement in Classification 405
9.7 Complete Exampie of Discriminant Analysis 407
9.7.1 Evaluation of Assumptions 407
9.7.1.1 Unequal Sample Sizes and Missing Data 407
9.7. I .2 Multivariate Normality 408
9.7.1.3 Linearity 408
9.7.1.4 Outliers 408
9.7.1.5 Homogeneity of Variance-Covariance Matrices 411
9.7.1.6 Multicollinearity and Singularity 41 1
9.7.2 Direct Discriminant Analysis 412
9.8 Comparison of Programs 430
9.8.1 SPSSPackage 430
9.8.2 SAS System 430
9.8.3 SYSTAT System 436
 CONTENTS xiii

I (1 Logistic Regression 437

10.1 General Purpose and Description 437
10.2 Kinds of Research Questions 439
10.2.1 Prediction of Croup Membership or Outcome 439
10.2.2 Importance of Predictors 439
10.2.3 Interactio~lsamong Predictors 440
10.2.4 Parameter Estimates 440
10.2.5 Classification of Cases 440
10.2.6 Significance of Prediction with Covariates 440
10.2.7 Effect Size 441
10.3 Limitations to Logistic Regression Analysis 441
. 10.3.1 Theoretical Issues 44 1
10.3.2 Practical Issues 442
10.3.2.1 Ratio of Cases to Variables 442
10.3.2.2 Adequacy of Expected Frequencies and Power 442
10.3.2.3 Linearity in the Logit 443
10.3.2.4 Absence of Multicollinearity 443
10.3.2.5 Absence of Outliers in the Solution 443
10.3.2.6 Independence of Errors 443
10.4 Fundamental Equations for Logistic Regression 444
10.4.1 Testing and Interpreting Coefficients 445
10.4.2 Goodness-of-Fit 446
10.4.3 Comparing Models 448
10.4.4 Interpretation and Analysis of Residuals 448
i0.4.5 Computer Analyse; of Sma!!-Samp!e Example 449
10.5 Types of Logistic Regression 453
10.5.1 Direct Logistic Regression 454
.n r 6
l u . 3 . ~Sequential Logistic Regressior, 454
10.5.3 Stadstical (Stepwise) Logistic Regression 454
10.5.4 Probit and Other Analyses 456
10.6 Some Important Issues 457
10.6.1 Statistical Inference 457
10.6.1.1 Assessing Goodness-of-Fit of Models 457
10.6.1.2 Tests of Individual Variables 459
10.6.2 Effect Size for a Model 460
10.6.3 Interpretation of Coefficients Using Odds 46 1
10.6.4 Coding Outcome and Predictor Categories 464
10.6.5 Number and Type of Outcome Categories 464
10.6.6 Classification of Cases 468
10.6.7 Hierarchical and Nonhierarchical Analysis 468
xiv CONTENTS

10.6.5 Importance of Predictor) 469

10.6.9 L O ~ I XReg~e\\ton
~IL' for Matched Group\ 469
10.7 Complete Examples of Logistic Regression 469
10.7.1 Evaluation of Limitations 470
10.7.1.1 Ratio of Cases to Variables and Missing Data 170
10.7.1.2 Multicollinearity 473
10.7.1.3 Outliers in the Solution 474
10.7.2 Direct Logistic Regression with Two-Category Outcome
and Continuous Predictors 474
10.7.2.1 Limitation: Linearity in the Logit 474
10.7.2.2 Direct Logistic Regression with Two-Category Outcome 474
10.7.3 Sequential Logistic Regression with Three Categories
of Outcome 481
10.7.3.1 Limitations of Multinomial Logistic Regression 481
10.7.3.2 Sequential Multinomial Logistic Regression 48 1
10.8 Comparisons of Programs 499
10.8.1 SPSS Package 499
10.8.2 SAS System 504
10.8.3 SYSTAT System 504

1I SurvivaWailure Analysis 506

11.1 General Purpose and Description 506
Kinds of Research Questions 507
1 1.2.1 Proportions Surviving at Various Times 507
1 !.2.2 Group Differences in Surviva! 508
1 i 2 . 3 Survivai Time with Covariates 508
1 1.2.3.1 Treatment Effects 508
1 1.2.3.2 Importance of Covar~ates 508
1 1.2.3.3 Parumeier Esrimates 508
1 1.2.3.4 Contingencies among Covariates 508
1 1.2.3.5 Effect Size and Power 509
11.3 Limitations to Survival Analysis 509
11.3.1 Theoretical Issues 509
11.3.2 Practical Issues 509
11.3.2.1 Sample Size and Missing Data 509
1 1.3.2.2 Normality of Sampling Distributions, Linearity,
and Homoscedasticity 5 10
1 1.3.2.3 Absence of Outliers 5 10
1 1.3.2.4 Differences between Withdrawn and
Remaining Cases 5 10
1 1.3.3.5 Change in Survival Conditions over Time 5 10
1 1.3.2.6 Proportionality of Hazards 5 10
1 1.3.2.7 Absence of Multicollinear~ty 5 10
 CONTENTS XV

11.4 Fundamental Equations for Survival 4nalysi5 51 1

1 1.4.1 L ~ f eTablej 511
i 1.4.2. Standard Error of Cumulat~vc:Proportion S u n i \ lng 5 13
1 1.4.3 Hazard and Density Funct1on5 5 14
1 1.4.4 Plot of Life Tables 5 15
1 1.4.5 Test for Group Differences 5 15
1 1.4.6 Computer Analyses of Small-Sample Example 5 17
11.5 Types of Survival Analyses 524
1 1.5.1 Actuarial and Product-Limit Life T ~ b l e s
and Survivor Functions 524
1 1.5.2 Prediction of Group Survival Times from Covariates 524
11 S.2.1 Direct, Sequential, and Statistical Analysis 527
11S.2.2 Cox Proportional-Hazards Model 527
1 1.5 2.3 Accelerated Failure-Time Models 529
1 1 S.2.4 Choosing a Method 535
11.6 Some Important Issues 535
1 1.6.1 Proportionality of Hazards 535
1 1.6.2 Censored Data 537
1 1.6.2.1 Right-Censored Data 537
1 1.6.2.2 Other Forms of Censoring 537
1 1.6.3 Effect Size and Power 538
11.6.4 Statistical Criteria 534
1 1.6.4.1 Test Statistics for Group Differences
in Survival Functions 539
1 1.6.4.2 Test Statistics for Prediction from Covariates 540
1 1.6.5 Predicting Survival Rate 540
1 1.6.5.1 Regression Coefticients (Parameter Estimates) 540
1 1.6.5.2 Odds Ratios 540
1 1.6.5.3 Expected Survival Rates 54 1
11.7 Complete Example of Survival Analysis 541
1 1.7.1 Evaluation of Assumptions 543
11.7.;.1 Accuracy of Input, Adequacy of Samplc Size, Missing Data,
and Distributions 543
1 1.7.1.2 Outliers 545
1 1.7.1.3 Differences between Withdrawn and Remaining Cases 549
I 1 7 1 4 Change in Survival Experience over Time 549
1 1.7.1.5 Proportionality of Hazards 549
1 1 7 1 6 Multicollinearity 55 1
1 1.7.2 Cox Regression Survival Analysis 55 1
1 1.7.2.1 Effect of Drug Treatment 552
1 1.7.2.2 Evaluation of Other Covariates 552
11.8 Comparison of Programs 559
i 1.8.1 SAS System 559
1 1.8.2 SPSS Package 559
1 1.8.3 SYSTAT System 566
xvi CONTENTS

1? Canonical Correlation 567

12.1 General Purpose and Description 567
12.2 Kinds of Research Questions 568
12.2.1 Number of Canonical Variate Pairs 568
12.2.2 Interpretation of Canonical Variates 569
12.2.3 Importance of Canonical Variates 569
12.2.4 Canonical Variate Scores 569
12.3 Limitations 569
12.3.1 Theoretical Limitations 569
12.3.2 Practical Issues 570
12.3.2.1 Ratio of Cases to IVs 570
12.3.2.2 Normality, Linearity, and Homoscedasticity 570
12.3.2.3 Missing Data 57 1
12.3.2.4 Absence of Outliers 57 1
12.3.2.5 Absence of Multicollinearity and Singularity 57 1
12.4 Fundamental Equations for Canonical Correlation 572
12.4.1 Eigenvalues and Eigenvectors 573
12.4.2 Matrix Equations 575
12.4.3 Proportions of Variance Extracted 579
12.4.4 Computer Analyses of Small-Sample Example 580
12.5 Some Important Issues 586
12.5.1 Importance of Canonical Variates 586
12.5.2 Interpretation of Canonical Variates 587
12.6 Complete Example of Canonical Correlation 587
12.6.1 Evaluation of Assumptions 588
12.6.1.1 Missing Data 588
12.6.1.2 Normality, Linearity, and Homoscedasticity 588
12.6.1.3 Outliers 591
12.6.1.4 Multicollinearity and Singularity 595
12.6.2 Canonical Correlation 595
12.7 Comparison of Programs 604
12.7.1 SAS System 604
12.7.2 SPSS Package 604
12.7.3 SYSTAT System 606

13 Principal Components and Factor Analysis 607

13.1 General Purpose and Description 607
13.2 Kinds of Research Questions 610
13.2.1 Number of Factors 610
 CONTENTS xvii

13.1 1 Nature of Factor. 61 1

13.2.3 Importance of Solut~on\and Factor\ hl 1
13.2.3 Testing Theory In FA 61 1
13.2.5 Estimating Scores on Factors 611
13.3 Limitations 611
13.3.1 Theoretical Issues 611
13.3.2 Practical Issues 6 12
13.3.2.1 Sample Size and Missing Data 613
13.3.2.2 Normality 613
13.3.2.3 Linearity 613
13.3.2.4 Absence of Outliers among Cases 61 3
13.3.2.5 Absence of Multicollinearity and Singularity 6 14
13.3.2.6 Factorability of R 6 14
13.3.2.7 Absence of Outliers among Variables 6 14
13.4 Fundamental Equations for Factor Analysis 615
13.4.1 Extraction 616
13.4.2 Orthogonal Rotation 620
13.4.3 Communalities, Variance, and Covariance 62 1
13.4.4 Factor Scores 622
L 3.4.5 Oblique Rotation 625
13.4.6 Computer Analyses of Small-Sample Example 628
13.5 Major Types of Factor Analyses 633
13.5.1 Factor Extraction Techniques 633
13.5.1.1 PCA vs. FA 634
13.5.1.2 Principal Components 635
13.5.1.3 Principal Factors 636
13.5.1.4 Image Factor Extraction 636
13.5.1.5 Maximum Likelihood Factor Extraction 636
13.5.1.6 Unweighted Least Squares Factoring 636
13.5.1.7 Generalized (Weighted) Least Squares Factoring 637
13.5.1.8 Alpha Factoring 637
13.5.2 Rotation 637
13.5.2.1 Orthogonal Rotation 638
13.5.2.2 Oblique Rotation 638
13.5.2.3 Geometric Interpretation 640
13.5.3 Some Practical Recommendations 642
13.6 Some Important Issues 643
13.6.1 Estimates of Communalities 643
13.6.2 Adequacy of Extraction and Number of Factors 644
13.6.3 Adequacy of Rotation and Simple Structure 646
13.6.4 Importance and Internal Consistency of Factors 647
13.6.5 Interpretation of Factors 649
13.6.6 Factor Scores 650
13.6.7 Comparisons among Solutions and Croups 65 1
xviii CONTENTS

13.7 Complete Example of FA 651

13.7.1 Evaluation of Limitations 652
13.7.1.1 Sample Size and Misjing Data 652
13.7.1.2 Normality 652
13.7.1.3 Linearity 652.
13.7.1.4 Outliers 652
13.7.1.5 Multicollinearity and Singularity 657
13.7.1.6 Outliers among Variables 657
13.7.2 Principal Factors Extraction with Varimax Rotation 657
13.8 Comparison of Programs 671
13.8.1 SPSS Package 674
13.8.2 SAS System 675
13.8.3 SYSTAT System 675

14 Structural Equation Modeling 676

14.1 General Purpose and Description 676
14.2 Kinds of Research Questions 680
14.2.1 Adequacy of the Model 680
14.2.2 Testing Theory 680
14.2.3 Amount of Variance in the Variables Accounted for
by the Factors 680
14.2.4 Reliability of the Indicators 680
14.2.5 Parameter Estimates 680
14.2.6 Intervening Variables 681
14.2.7 Group Differences 68 1
14.2.8 Long~tudinalDifferences 68 1
14.2.9 Multilevel Modeling 68 1
14.3 Limitations to Structural Equation Modeling 682
i4.3. i Tlieoreiicai issues 682
14.3.2 Practical Issues 682
14.3.2.1 Sample Size and Missing Data 682
14.3.2.2 Multivariate Normality and Absence of Outliers 683
14.3.2.3 Linearity 683
14.3.2.4 Absence of Multicollinearity and Singularity 683
14.3.2.5 Residuals 684
14.4 Fundamental Equations for Structural Equations Modeling 684
14.4.1 Covariance Algebra 684
14.4.2 Model Hypotheses 686
14.4.3 Model Specification 688
i4.4.4 Model Estimation 690
14.4.5 Model Evaluation 694
14.1.6 Computer Analysis of Small-Sample Example 696
 CONTENTS xix

14.5 Some Important Issues 709

14.5.1 Model Identification 709
14.5.2 Estimation Techniques 7 13
14.5.2.1 Estimation Methods and Sample Size 714
14.5.2.2 Estimation Methods and Nonnormality 7 14
14.5.2.3 Estimation Methods and Dependence 7 15
14.5.2.4 Some Recommendations for Choice
of Estimation Method 7 15
14.5.3 Assessing the Fit of the Model 715
14.5.3.1 Colnpwative Fit Indices 7 16
14.5.3.2 Absolute Fit Index 7 18
14.5.3.3 Indices of Proportion of Variance Accounted 7 18
14.5.3.4 Degree of Parsimony Fit Indices 7 19
14.5.3.5 Residual-Based Fit Indices 720
14.5.3.6 Choosing among Fit Indices 720
14.5.4 Model Modification 72 1
14.5.4.1 Chi-Square Difference Test 72 1
14.5.4.2 Lagrange Multiplier (LM) Test 72 1
14.5.4.3 Wald Test 723
14.5.4.4 Some Caveats and Hints on Model Modification 728
14.5.5 Reliability and Proportion of Variance 728
14.5.6 Discrete and Ordinal Data 729
14:5:7 Multiple Group Models 730
14.5.8 Mean and Covariance Structure Models 73 1
14.6 Complete Examples of Structural Equation
Modeling Analysis 732
14.6.1 Confirmatory Factor Analysis of the WISC 733
14.6.1.1 Model Specification for CFA 7 72
14.6.1.2 Evaluation of Assumptions for CFA 733
14.6.1.3 CFA Model Estimation and Preliminary Evaluation 734
14.6.1.4 Model Modification 743
14.6.2 SEM of Health Data 750
14.6.2.1 SEM Model Specification 750
14.6.2.2 Evaluation of Assumptions for SEM 75 1
14.6.2.3 SEM Model Estimation and Preliminary Evaluation 755
14.6.2.4 Model Modification 759
14.7 Comparison of Programs 773
14.7.1 EQS 773
14.7.2 LISREL 773
14.7.3 AMOS 780
14.7.4 SAS System 780

13 Multilevel Linear Modeling 781

15.1 General Purpose and Description 781
CONTENTS

15.2 Kinds of Research Questions 784

15.2.1 GSOLIP Differences in Means 784
15.2.2 Groiip Differences in Slopes 784
15.2.3 Cross-Level Interactions 785
15.2.4 Meta-Analysis 785
15.2.5 Relative Strength of Predictors at Various Levels 785
15.2.6 Individual and Group Structure 785
15.2.7 Path Analysis at Individual and Group Levels 786
15.2.8 Analysis of Longitudinal Data 786
15.2.9 Multilevel Logistic Regression 786
15.2.10 Multiple Response Analysis 786
15.3 Limitations to Multilevel Linear Modeling 786
15.3.1 Theoretical Issues 786
. 15.3.2 Practical Issues 787
15.3.2.1 Sample Size, Unequal-n, and Missing Data 787
15.3.2.2 Independence of Errors 788
15.3.2.3 Absence of Multicollinearity and Singularity 785,
15.4 Fundamental Equations 789
15.4.1 Intercepts-Only Model 792
15.4.I. l The Intercepts-Only Model: Level- 1 Equation 793
15.4.1.2 The Intercepts-Only Model: Level-2 Equation 793
15.4.1.3 Conlputer Analysis of Intercepts-only Model 794
15.4.2 Model with a First-Level Predictor 799
15.4.2.1 Level- l Equation for a Model with a Level- 1 Predictor 799
15.4.2.2 Level-2 Equations for a Model with a Level- I Predictor 801
15.4.2.3 Computer Analysis of a Model with a Level-1 Predictor 802
15.4.1 Model with Predictors at First and Second Levels X07
15.4.3.1 Level-1 Equation for Model with Predictors
at Both Levels 807
15.4.3.2 Level-2 Equations for Model with Predictors
at Both level< 807
15.4.3.3 Computer Analyses of Model with Predictors
at First and Second Levels 808
15.5 Types of M L M 814
15.5.1 Repeated Measures 8 14
15.5.2 Higher-Order MLM 8 19
15.5.3 Latent Variables 8 19
15.5.4 Nonnormal Outcome Variables 820
15.5.5 Multiple Response Models 82 1
15.6 Some Important Issues 822
15.6.1 Intraclass Correlation 822
15.6.2 Centering Predictors and Changes in Their Interpretations 823
15.6.3 Interactions 826
15.6.4 Random and Fixed Intercepts and Slopes 826
 CONTENTS xxi

15.6.5 Statistical Inference 810

15.6.5.1 Assessing Models 830
15.6.5.2 Test5 of In&\ idual El'fects 63 1
15.6.6 Effect Size 832
15.6.7 Estimation Techniques and Convergence Problems 833
15.6.8 Exploratory Model Building 834
,15.7 Complete Example of MLM 835
15.7. I Evaluation of Assumptions 835
15.7. I. 1 Sample Sizes, Missing Data, and Distributions 835
15.7.1.2 Outliers 838
15.7.1.3 Multicollinearity and Singularity 839
15.7.1.4 Independence of Errors: Intraclass Correlations 839
15.7.2 Multilevel Modeling 840
15.8 Comparison of Programs 852
15.8.1 SAS System 852
15.8.2 SPSS Package 856
!5.8.3 HLM Prograrr. 856
15.8.4 MLwiN Program 857
15.8.5 SYSTAT System 857

16 Multiway Frequency Analysis 858

16.1 General Purpose and Description 858
16.2 Kinds of Research Questions 859
16.2.1 Associations among Variables 859
16.2.2 Effect on a Dependent Variable 860
16.2.3 Parameter Estimates 860
16.2.4 Importance of Effects 860
16.2.5 Effect Size 860
16.2.6 Specific Comparisons and Trend Analysis 860
16.3 Limitations to Multiway Frequency Analysis 861
16.3.1 Theoretical Issues 86 1
16.3.2 Practical Issues 861
16.3.2.I Independence 86 1
16.3.2.2 Ratio of Cases to Variables 861
16.3.2.3 Adequacy of Expected Frequencies 862
16.3.2.4 Absence of Outliers in the Solution 863
16.4 Fundamental Equations for Multiway Frequency Analysis 863
16.4.1 Screening for Effects 864
16.4.1.1 Total Effect 865
16.3.1.2 First-Order Effects 866
16.4 1 3 Second-Order Effects 867
16.4.1.3 Th~rd-OrderEffect 87 I
xxii CONTENTS

16.4.1 blodel~ng 87 1
16.4..3 Evaluatic~iand Interpretation 574
16.4.3.i Rcsiduais 87-4
16.4.3.2 Parameter Estimates 874
16.4.4 Cotnputer Analyses of Small-Sample Example 880
16.5 Some Important Issues 887
16.5.1 Hierarchical and Nonhierarchical Models 887
16.5.2 Statistical Criteria 888
16.5.2.1 Tests of Models 888
16.5.2.2 Tests of Individual Effects 888
16.5.3 Strategies for Choosing a Model 889
16.5.3.1 SPSS HILOGLINEAR (Hierarchical) 889
16.5.3.2 SPSS GENLOG (General Log-Linear) 889
16.5.3.3 SAS CATMOD and SPSS LOGLINEAR
(General Log-Linear) 890
16.6 Complete Example of Multiway Frequency Analysis 890
16.6.1 Evaluation of Assumptions: Adequacy
of Expected Frequencies 890
16.6.2 Hierarchical Log-Linear Analysis 89 1
16.6.2.1 Preliminary Model Screening 89 1
16.6.2.2 Stepwise Model Selection 893
16.6.2.3 Adequacy of Fit 895
16.6.2.4 Interpretation of the Selected Model 901
16.7 Comparison of Programs 908
16.7.1 SPSS Package 911
16.7.2 SAS System 9 12
16.7.3 SYSTAT System 9!3

17 An Overview of the General Linear Model 913

17.1 Linearity and the General Linear Model 913
17.2 Bivariate to Multivariate Statistics and Overview
of Techniques 913
17.2.1 BivariateForm 913
17.2.2 Simple Multivariate Form 914
17.2.3 Full Multivariate Form 9 17
17.3 Alternative Research Strategies 918

I8 Time-Series Analysis (available online at

www.ablongman.com/tabachnick5e) 18-1
18.1 General Purpose and Description 18-1
 CONTENTS xxiii

18.2 Kinds of Research Questions 18-3

18.2.1 Pattern of Autocorrelation 18-5
18.2.2 Seasonal Cycles and Trends 18-5
18.2.3 Forecasting 18-5
18.2.4 Effect of an Intervention 18-5
18.2.5 Comparing Time Series 18-5
18.2.6 Time Series with Covariates 18-6
18.2.7 Effect Size and Power 18-6
' 18.3 Assumptions of Time-Series Analysis 18-6
18.3.1 Theoretical Issues 18-6
18.3.2 Practical Issues 18-6
18.3.2.1 Normality of Distributions of Residuals 18-6
18.3.2.2 Homogeneity of Variance and Zero Mean of Residuals 18-7
18.3.2.3 Independence of Residuals 18-7
18.3.2.4 Absence of Outliers 18-7
18.4 Fundamental Equations for Time-Series ARIMA Models 18-7
18.4.1 Identification ARIMA (p, d, q) Models 18-8
18.4.1.1 Trend Components, d: Making the Process Stationary 18-8
18.4.1.2 Auto-Regressive Components 18-1 1
18.4.1.3 Moving Average Components 18-12
18.4.1.4 Mixed Models 18-13
18.4.1.5 ACFs and PACFs 18-13
18.4.2 Estimating Model Parameters 18- 16
18.4.3 Diagnosing a Model 18- 19
18.4.4 Computer Analysis of Small-Sample
Time-Series Example 18- 19
18.5 Types of Time-Series Analyses 18-27
18.5.1 Models with Seasonal Components 18-27
18.5.2 Models with Interventions 18-30
18.5.2.1 Abrupt. Permanent Effects 18-32
18.5.2.2 Abrupt, Temporary Effects 18-32
18.5.2.3 Gradual, Permanent Effects 18-38
18.5.2.4 Models with Multiple Interventions 18-38
18.5.3 Adding Continuous Variables 18-38
18.6 Some Important issues 18-41
18.6.1 Patterns of ACFs and PACFs 18-4 1
18.6.2 Effect Size 18-44
18.6.3 Forecasting 18-45
18.6.4 Statistical Methods for Comparing Two Models 18-45
18.7 Complete Example of a Time-Series Analysis 18-47
18.7.1 Evaluation of Assumptions 18-48
18.7.1.1 Normality of Sampling Distributions 18-48
18.7.1.2 Homogeneity of Variance 18-48
18.7.1.1 Outliers 18-48
xxiv CONTENTS

1 8.7 .'7 Baseline ivlodel identification and Estiination 18-48

I Y .7.3 B a s l i n e Model Diagnosis 18-19
18.7.1 Intervention Analysis 18-55
18.7.4.1 Model Diagnosis 18-55
18.7.4.2 Model Interpretation 18-56
18.8 Comparison of Programs 18-60
18.8.1 SPSS Package 18-61
18.8.2 SAS System 18-61
18.8.3 SYSTAT System 18-61

Appendix A A Skimpy Introduction to Matrix Algebra 924

A.1, The Trace of a Matrix 925
A.2 Addition or Subtraction of a Constant to a Matrix 925
A.3 Multiplication or Division of a Matrix by 3 Constant 925
A.4 Addition and Subtraction of Two Matrices 926
A.5 Multiplication, Transposes, and Square Roots of Matrices 927
A.6 Matrix "Division" (Inverses and Determinants) 929
A.7 Eigenvalues and Eigenvectors: Procedures
for Consolidating Variance from a Matrix 930

Appendix K Research Designs for Complete Examples 934

B.l Women's Health and Drug Study 934
B.2 Sexual Attraction Study 935
B.3 Le'lrning Disabilities Data Bank 938
B.4 Reaction Time to Identify Figures 939
B.5 Field Studies of Noise-Induced Sleep Disturbance 939
B.6 Clinical Trial for Primary Biliary Cirrhosis 940
B.7 Impact of Seat Belt Law 940

Appendix C Statistical Tables 941

C.l Normal Curve Areas 942
C.2 Critical Values of the t Distribution for a = .05 and .01,
Two-Tailed Test 943
 CONTENTS YYV

C.3 Critical Values of the F Distribution 944

C.1 Critical Values of Chi Square ( X ' ) 949
C.5 Critical Values for Squared ICIultiple Correlation ( R ~in) Forward
Stepwise Selection 950
C.6 Critical Values for F,,,, ( s ~ , , , / s ~Distribution
~,,) for a = .05
and .O1 952

References 953

Index 963
 PREFACE

Obesity threatened, and we've had to consider putting the book on a diet. We've added only'one
chapter this time around, Multilevel Linear Modeling (Chapter 15). and some spiffy new techniques
for dealing with missing data (in Chapter 4). Otherwise, we've mostly streamlined and said good-
bye to some old friends. We've forsaken the Time-Series Analysis chapter in the text, but you'll be
able to download it from the publisher's web site at www.ablongman.com/tabachnick5e. Another
sadly forsaken old friend is SYSTAT. We still love the program, however, for its right-to-the-point
analyses and terrific graphics, and are pleased that most of the graphics have been incorporated into
SPSS. Although absent from demonstrations, features of SYSTAT, and any other programs we've
cut, still appear in the last sections of Chapters 5 through 16, and in online Chapter 18, where pro-
grams are compared. We've changed the order of some chapters: canonical correlation seemed rather
difficult to appear as early as it did, and survival analysis seemed to want to snuggle up to logistic
regression. .Act~lally,the order doesn't seem to matter much: perusal of syllabi on the Web convinces
us that professors feel free to present chapters in any order they choose-and that's fine with us.
Multilevel linear modeling (MLM) seems to have taken the world by storm; how did we ever
live without it? Real life is hierarchical-students come to us within classrooms, teachers work
within different schools, patients share wards and nursing staff, and audiences attend different per-
formances. We hardly ever get to break these groups apart for research purposes, so we have to deal
with intact groups and all their shared experiences. MLM lets us do this without violating all of the
statistical assumptions we learned to know and hate. Now that SAS and SPSS can deal with these
models, we're ready to tackle the real world. Hence, a new chapter.
SAS and SPSS also now offer reasonable ways to impute illissing data through multiple-
imputation techniques and fii_llly assess miscing data patterns, respectively. We expanded Chapter 4
to detnonstrate these enhancements. SPSS and SAS keep adding goodies, which we'll try to show
off. As before, we adapt our syntax from Windows menus whenever possible, and all of our data sets
are available on the book's web page (www.ablongman.com/tabachnick5e). We've also paid more
attention to effect sizes and, especially, confidence intervals around effect sizes. Michael Simpson of
[he Austraiian Nationai University has kindiy given us permission to include some nifty SPSS and
SAS syntax and data files in our web page downloads. Jim Steiger and Rachel Fouladi have gra-
ciously given us permission to include their DOS program that finds confidence intervals around R?
One thing we'll never change is our practical bent, focusing on the benefits and lirriitations of
applications of a technique to a data set-when, why, and how to do it. The math is wonderful, and
we suggest (but don't insist) that students follow along through section four of each chapter using
readily available software for matrix manipulations or spreadsheets. But we still feel that under-
standing the math is not enough to insure appropriate analysis of data. And our readers assure us that
they really are able to apply the techniques without a great deal of attention to the math of section
four. Our small-sample examples remain silly; alas, our belly dancing days are over. As for our most
recent reviewers, kindly provided by our publisher, we had the three bears checking out beds: too
hard, too soft, and just right. So we've not changed the tone or level of difficulty.
Some extremely helpful advice wax offered by S t e ~ eOsterlincl of the Univerxity of
Missouri-Columbia and Jeremy Jewel of Southern Illinois University-Edwal-dsville. We also

xxvii
xxviii PREFACE

heartily thank Lisa Harlow of the Un~versityof Rhodr: I.\land. who wrote an extenhibe. in\i?htful
E ~ I I L I IMI ~o I~ Ii ~ l i ~inz 7002.
review of the entire fourth edition of 011s book in Sti.~l(~t~[i.lll g LVr asain
thank the reviewers of earlier editions of our book, but fears of breaking the backs of current students
dissuade us from listing them all once more. You know who you are; we still care. Our thanks to the
reviewers of this edition: Joseph Benz, University of Nebraska-Kearney; Stanley Cohen, West Vir-
ginia University; Michael Granaas, University of South Dakota; Marie Hammond. Tennessee State
University at ~ a s h v i l l eJosephine
; Korchmaros, Southern Illinois University; and Scott Roesch, San
Diego State University.
As always, the improvements are largely due to reviewers and those colleagues who have taken
the time to email us with suggestions and corrections. Any remaining errors and lack of clarity are
due to us alone. As always, we hope the book provides a few smiles as well as help in analyzing data.

Barbara G. Tabachnick
Linda S. Fidell
 CHAPTER

Introduction

1 . Multivariate Statistics: Why?

Multivariate statistics are increasingly popular techn~quesused for anaiyzing compiicated data sets.
They provide analysis when there are many independent variables (IVs) and/or many dependent
variables (DVs), all correlated with one another to varying degrees. Because of the difficulty of
addressing complicated research questions with univariate analyses and because of the availability
of canned software for performing multivariate analyses, multivariate statistics have become widely
used. Indeed, a standard univariate statistics course only begins to prepare a student to read research
literature or a researcher to produce it.
But how much harder are the multivariate techniques? Compared with the multivariate meth-
ods, univariate statistical methods are so straightforward and neatly structured that it is hard to
believe they once took so much effort to master. Yet many researchers apply and correctly interpret
results of intricate analysis of variance before the grand structure is apparent to them. The same can
be true of multivariate statistical methods. Although we are delighted if you gain insights into the full
multivariate general linear model,( we have accomplished our goal if you feel comfortable selecting
and setting up multivariate analyses and interpreting the computer output.
Multivariate methods are more complex than univariate by at least an order of magnitude. But
for the most part, the greater complexity requires few conceptual leaps. Familiar concepts such as
sampling distributions and homogeneity of variance simply become more elaborate.
Multivariate models have not gained popularity by accident--or even by sinister design. Their
growing popularity parallels the greater complexity of contemporary research. In psychology, for
example, we are less and less enamored of the simple, clean, laboratory study in which pliant, first-
year college students each provides us with a single behavioral measure on cue.

1.1.1 The Domain of Multivariate Statistics: Numbers of IVs

and DVs
Multi~ariatcstatistical methods are an extension of univariate and bivariate statistics. Multivariate
statistics are the cnmplere or general case, whereas univariate and bivariate statistics are cpecial cases

'Chapter 17 attempts to foster such insights

of the multivariate model. If your desisn ha.; many ~ast:tble\.rnultlvar~:~te techn~clue\often Ict ~ O L I
perform a single analysis instead oft: series of univari~iteor bi\.asinte anaiyses.
Variables are roughly dichoton~izedinto two major types-independent and dependent. Inde-
pendent variables (IVs) are the differing conditions (treatment vs. placebo) to which you expose your
subjects, or characteristics (tall or short) that the subjects themselves bring into the research situa-
tion. IVs are usually considered predictor variables because they predict the DVs-the response or
outcome variables. Note that IV and DV are defined within a research context; a DV in one research
setting may be an IV in another.
Additional t e k s for IVs and DVs are predictor-criterion, stimulus-response, task-performance,
or simply input-output. We use IV and DV throughout this book to identify variables that belong on
one side of an equation or the other, without causal implication. That is, the terms are used for conve-
nience rather than to indicate that one of the variables caused or determined the size of the other.
The term univariclte statistics refers to analyses in which there is a single DV. There may be,
however, more than one IV. For example, the amount of social behavior of graduate students (the
DV) is studied as a function of course load (one IV) and type of training in social skills to which stu-
dents are exposed (another IV). Analysis of variance is a commonly used univariate statistic.
Blviirtaie siiiilsiies fieqileniiy refers io anaiysis or' two variabies where neither is an experi-
mental IV and the desire is simply to study the relationship between the variables (e.g., the relation-
ship between income and amount of education). Bivariate statistics, of course, can be applied in an
experimental setting, but usually they are not. Prototypical examples of bivariate statistics are the
Pearson product-moment correlation coefficient and chi square analysis. (Chapter 3 reviews univari-
ate and bivariate statistics.)
With multivariate statistics, you simultaneously analyze multiple dependent and multiple inde-
pendent variables. This capability is important in both nonexperimental (correlational or survey) and
experimental research.

1.1.2 Experimental and Nonexperimental Research

A critical distinction between experimental and nonexperimental research is whether the researcher
manipulates the levels of the IVs. In an experiment, the researcher has control over the levels (or con-
ditions) of at least one IV to which a subject is exposed by determining what the levels are, how they
are implemented, and how and when cases are assigned and exposed to them. Further, the experi-
menter randomly assigns subjects to levels of the IV and controls all other influential factors by hold-
ing them constant, counterbalancing, or randomizing their influence. Scores on the DV are expected to
be the same, within random variation, except for the influence of the IV (Campbell & Stanley. 1966).
If there are systematic differences in the DV associated with levels of the IV, these differences are
attributed to the IV.
For example, if groups of undergraduates are randomly assigned to the same material but dif-
ferent types of teaching techniques, and afterward some groups of undergraduates perform better
than others, the difference in performance is said, with some degree of confidence. to be caused by
the difference in teaching technique. In this type of research, the terms independent and dependent
have obvious meaning: the value of the DV depends on the manipulated level of the IV. The IV is
manipulated by the experimenter and the score on the DV depends on the level of the IV.
 In noneuperimental !correlational or wrvey) rewarcli. the le\els o f the IV(s) are not manipu-
lated by the researcher. The researcher can Jttfi~iethe IV. but has no c~titrolO L ~ the K ahsignment of
subjects to levels of it. For example, groups of people may be categorized into geographic area of res-
idence (Northeast, Midwest, e t ~ . )but
, only the definition of the variable is under researcher control.
Except for the military or prison, place of residence is rarely subject to manipulation by a researcher.
Nevertheless, a naturally occurring difference like this is often considered an IV and is used to pre-
dict some other b~nex~eriinental (dependent) variable such as income. In this type of research. the
distinction between IVs and DVs is usually arbitrary and many researchers prefer to call IVs predic-
tors and DVs criterion variables.
In nonexperimental research, it is very difficult to attribute causality to an IV. If there is a sys-
tematic difference in a DV associated with levels of an IV, the two variables are said (with some
degree of confidence) to be related, but the cause of the relationship is unclear. For example, income
as a DV might be related to geographic area, but no causal association is implied.
Nonexperimental research takes many forms, but a common example is the survey. Typically,
many people are surveyed, and each respondent provides answers to many questions, producing a
large number of variables. These variables are usually interrelated in highly complex ways, but uni-
variate and bivariate statistics are not sensitive to this complexity. Bivariate correlations between aii
pairs of variables, for example, could not reveal that the 20 to 25 variables measured really represent
only two or three "supervariables."
Or, if a research goal is to distinguish among subgroups in a sample (e.g., between Catholics
and Protestants) on the basis of a variety of attitudinal variables, we could use several univariate t
tests (or analyses of variance) to examine group differences on each variable separately. But if the
variables are related, which is highly likely, the results of many t tests are misleading and statistically
suspect.
With the use of multivariate statistical techniques, complex interrelationships among variables
are revealed and assessed in statistical inference. Further, it is possible to keep the overall Type I
error rate at, say, 5%. no matter how many variables are tested.
Although most multivariate techniques were developed for use in nonexperimental research,
they are also useful in experimental research in which there may be multiple IVs and multiple DVs.
With multiple IVs, the research is usually designed so that the IVs are independent of each other and
a straightforward correction for numerous statistical tests is available (see Chapter 3). With multiple
DVs, a problem of inflated error rate arises if each DV is tested separately. Further, at least some of
the DVs are likely to be correlated with each other, so separate tests of each DV reanalyze some of
the same variance. Therefore, multivariate tests are used.
Experimental research designs with multiple DVs were unusual at one time. Now, however,
with attempts to make experimental designs more realistic, and with the availability of computer pro-
grams, experiments often have several DVs. It is dangerous to run an experiment with only one DV
and risk missing the impact of the IV because the most sensitive DV is not measured. Multivariate
statistics help the experimenter design more efficient and more realistic experiments by allowing
measurement of multiple DVs without violation of acceptable levels of Type I error.
One of the few considerations not relevant to choice of stnti.rtica1 technique is whether the data
are experimenta! or correlational. The statistical methods "work" whether the researcher manipulated
the levels of the IV. But attribution of causality to results is cn~ciallyaffected by the experimental-
nonexperimental distinction.
4 CHAPTER I

1.1.3 Computers and hlultivariate Statistics

One answer to the question "Why multivariate statistics'?" is that the techniques are now accessible
by computer. Only the most dedicated number cruncher would consider doing real-life-sized prob-
lems in multivariate statistics without a computer. Fortunately, excellent multivariate programs are
available in a number of computer packages.
Two packages are demonstrated in this book. Examples are based on programs in SPSS (Sta-
tistical Package for the Social Sciences) and SAS.
If you have hccess to both packages, you are indeed fortunate. Programs within the packages
do not completely overlap, and some problems are better handled through one package than the
other. For example, doing several versions of the same basic analysis on the same set of data is par-
ticularly easy with SPSS whereas SAS has the most extensive capabilities for saving derived scores
from data screening or from intermediate analyses.
Chapters 5 through 16 (the chapters that cover the specialized multivariate techniques) and Chap-
ter 18 (available at www.ablongman.com/tabachnick5e) offer explanations and illustrations of a variety
of programs2 within each package and a comparison of the features of the programs. We hope that once
you understand the techniques, you will be able to generalize to virtually any mu!tivxlriate program.
Recent versions of the programs are implemented in Windows, with menus that implement
most of the techniques illustrated in this book. All of the techniques may be implemented through
syntax, and syntax itself is generated through menus. Then you may add or change syntax as desired
for your analysis. For example, you may "paste" menu choices into a syntax window in SPSS, edit
the resulting text, and then run the program. Also, syntax generated by SPSS menus is saved in the
"journal" file (spss.jn1) which also may be accessed and copied into a syntax window. Syntax gener-
ated by SAS menus is recorded in a "log" file. The contents may then be copied to an interactive win-
dow, edited, and run. Do not overlook the help files in these programs. Indeed, SAS and SPSS now
provide the entire set of user manuals on CD, often with more current information than is available
in printed manuals.
Our demonstrations in this book are based on syntax generated through menus whenever fea-
sible. We would love to show you the sequence of menu choices, but space does not permit. And, for
the sake of parsimony, we have edited program output to illustrate the material that we feel is the
most important for interpretation. We have also edited out some of the unnecessary (because it is
default) syntax that is generated through menu choices.
With commercial computer packages, you need to know which version of the package you are
using. Programs are continually being changed, and not all changes are immediately implemented at
each facility. Therefore, many versions of the various programs are simultaneously in use at differ-
ent institutions; even at one institution, more than one version of a package is sometimes available.
Program updates are often corrections of errors discovered in earlier versions. Occasionally,
though, there are major revisions in one or more programs or a new program is added to the package.
Sometimes defaults change with updates, so that output looks different although syntax is the same.
Check to find out which version of each package you are using. Then be sure that the manual you are
using is consistent with the version in use at your facility. Also check updates for error correction in
previous releases that may be relevant to some of your previous runs.
Except where noted, this book reviews Windows versions of SPSS Version 13 and SAS Ver-
sion 9.1. Information on availability and versions of software, macros, books, and the like changes
almost daily. We recommend the Internet as a source of "keeping up."
'We have retained descriptions of features of SYSTAT in these sections despite the removal of detailed demonstration\ of that
program in this edition.
1.14 Garbage In, Roses Out?
The trick in multivariate statistics is not in computation; that is easily done as discu~sedabove. The
trick is to select reliable and valid measurements, choose the appropriate program, use it correctly,
and know how to interpret the output. Output from commercial computer programs, with their beau-
tifully formatted tables, graphs, and matrices, can make garbage look like roses. Throughout'this
book, we try to suggest clues that reveal when the true message in the output more closely resembles
the fertilizer than the flowers.
Second, when you use multivariate statistics, you rarely get as close to the raw data as you do
when you apply univariate statistics to a relatively few cases. Errors and anomalies in the data that
would be obvious if the data were processed by hand are less easy to spot when processing is entirely
by computer. But the computer packages have programs to graph and describe your data in the sim-
plest univariate terms and to display bivariate relationships among your variables. As discussed in
Chapter 4, these programs provide preliminary analyses that are absolutely necessary if the results of
multivariate programs are to be believed.
There are also certain costs associated with the benefits of using multivariate procedures. Ben-
efits of increased flexibility in research design, for instance, are sometimes paralleled by increased
ambiguity in interpretation of results. In addition, multivariate results can be quite sensitive to which
analytic strategy is chosen (cf. Section 1.2.4) and do not always provide better protection against sta-
tistical errors than their univariate counterparts. Add to this the fact that occasionally you still cannot
get a firm statistical answer to your research questions, and you may wonder if the increase in com-
plexity and difficulty is warranted.
Frankly, we think it is. Slippery as some of the concepts and procedures are, these statistics
provide insights into relationships among variables that may more closely resemble the complexity
of the "real" world. And sometimes you get at least partial answers to questions that could not be
asked at all in the univariate framework. For a complete analysis, making sense of your data usually
requires a judicious mix of multivariate and univariate statistics.
And the addition of multivariate statistical methods to your repertoire makes data analysis a lot
more fun. If you liked univariate statistics, you will love multivariate statistic^!^

1.2 Some Useful Definitions

In order to describe multivariate statistics easily, it is useful to review some common terms in research
design and basic statistics. Distinctions were made in preceding sections between IVs and DVs &id
between experimentai and nonexperimental research. Additienal terms that are encountered repeat-
edly in the book but not necessarily related to each other are described in this section.

1.2.1 Continuous, Discrete, and Dichotomous Data

In applying statistical techniques of any sort, it is important to consider the type of measurement and
the nature of the correspondence between numbers and the events that they represent. The distinction
made here is among continuous, discrete, and dichototnous variables; you may prefer to substitute
the terms intrrvcll or y~l~lntitutivr
for cot7ritluo~r.\and tlortzinr~l.~ , ~ z t ~ g o tor for (iirC1oto-
- i cq~inlitntil'~
~~l
mnus and discrete.

' ~ o n ' teven think about it.

6 CHAPTER I

Conti~?uousvariable\ are measured on a ccale that changes calues m~oothlkrather than II! step\.
Continuous variables take on any u l u e within the range of the hcale. and the \ i ~ eof the number
retlects the amount of the variable. Precision is limited by the measuring instrument, not by the nature
of the scale itself. Some examples of continuous variables are time as measured on an old-fashioned
analog clock face. annual income, age, temperature, distance, and grade point average (GPA). .

Discrete variables take on a finite and usually small number of values, and there is no smooth
transition from one value or category to the next. Examples include time as displayed by a digital
clock, continents, categories of religious affiliation, and type of community (rural or urban).
Sometimes aiscrete variables are used in multivariate analyses as if continuous if there are
numerous categories and the categories represent a quantitative attribute. For instance, a variable that
represents age categories (where, say, 1 stands for 0 to 4 years, 2 stands for 5 to 9 years, 3 stands for
10 to 14 years, and so on up through the normal age span) can be used because there are a lot of cat-
egories and the numbers designate a quantitative attribute (increasing age). But the same numbers
used to designate categories of religious affiliation are not in appropriate form for analysis with
many of the techniques%ecause religions do not fall along a quantitative continuum.
Discrete variables composed of qualitatively different categories are sometimes analyzed after
being changed into a number of dichotomous or two-!eve! variables (e.g., Catholic vs. son-Cath~lic,
Protestant vs. non-Protestant, Jewish vs. non-Jewish, and so on until the degrees of freedom are
used). Recategorization of a discrete variable into a series of dichotomous ones is called dummy vari-
able coding. The conversion of a discrete variable into a series of dichotomous ones is done to limit
the relationship between the dichotomous variables and others to linear relationships. A discrete
variable with more than two categories can have a relationship of any shape with another variable,
and the relationship is changed arbitrarily if assignment of numbers to categories is changed.
Dichotomous variables, however, with only two points, can have only linear relationships with other
variables; they are, therefore, appropriately analyzed by methods using correlation in which only lin-
ear relationships are analyzed.
The distinction between continuous and discrete variables is not always clear. If you add
enough digits to the digital clock, for instance, it becomes for all practical purposes a continuous
measuring device, whereas time as measured by the analog device can also be read in discrete cate-
gories such as hours or half hours. In fact, any continuous measurement may be rendered discrete (or
dichotomous) with some loss of information, by specifying cutoffs on the continuous scale.
The property of variables that is crucial to application of multivariate procedures is not the type
of measurement so much as the shape of distribution, as discussed in Chapter 4 and elsewhere. Non-
normally distributed continuous variables and dichotomous variables with very uneven splits
between the categories present problems to several of the multivariate analyses. This issue and its
resolutior, are disciissed at some length in Chapter 4.
Another type of measurement that is used sometimes produces a rank order (ordinal) scale.
This scale assigns a number to each subject to indicate the subject's position vis-8-vis other subjects
along some dimension. For instance, ranks are assigned to contestants (first place, second place,
third place, etc.) to provide an indication of who was best-but not by how much. A problem with
ordinal measures is that their distributions are rectangular (one frequency per number) instead of
normal. unless tied ranks are permitted and they pile up in the middle of the distribution.

'some tnultivariate techliiq~~es

ie.g.. logi\tic regres.;ion, S E M ) are clpproprlate for all type, o t var~ables
 Introduction 7

In practice. we often treat variables as ~ tthey

' are continuous when the underlying scale is
thought to be continuous but the measured scale actually 1s ordinal. the number of categories is
large-say, seven or more-and the data meet other assumptions of the analysis. For instance, the
number of correct items on an objective test is technically not continuous because fractional values
are not possible, but it is thought to measure some underlying continuous variable such as course
mastery. Another example of a variable with ambiguous measurement is one measured on a Likert-
type scale in which consumers rate their attitudes toward a product as "strongly like," "moderately
like," "mildly like,'' "neither like nor dislike," "mildly dislike," "moderately dislike," or "strongly
dislike." As mentioned previously, even dichotomous variables may be treated as if continuous under
some conditions. Thus, we often use the term "contirz~ious"throughout the remainder of this book
whether the measured scale itself is continuous or the variable is to be treated as if continuous. We
use the term "discrete" for variables with a few categories, whether the categories differ in type or
quantity.

1.2.2 Samples and Populations

Samples are measured in order to make generaiizatiotib about popillations. Ideal!y, samp!es 2re
selected, usually by some random process, so that they represent the population of interest. In real
life, however, populations are frequently best defined in terms of samples, rather than vice versa; the
population is the group from which you were able to randomly sample.
Sampling has somewhat different connotations in nonexperimental and experimental research.
In nonexperimental research, you investigate relationships among variables in some predefined pop-
ulation. Typically, you take elaborate precautions to ensure that you have achieved a representa-
tive sample of that population; you define your population: then do your best to randomly sample
from it.5
In experimental research, you attempt to create different populations by treating subgroups
from an originally homogeneous group differently. The sampling objective here ih to ensure that all
subjects come from the same population before you treat them differently. Random sampling con-
sists of randomly assigning subjects to treatment groups (levels of the IV) to ensure that, before dif-
ferential treatment, all subsamples come from the same population. Statistical tests provide evidence
as to whether, after treatment, all samples still come from the same population. Generalizations
about treatment effectiveness are made to the type of subjects who participated in the experiment.

1.2.3 Descriptive and Inferential Statistics

Descriptive statistics describe samples of subjects in terms of variables or combinations of variables.
Inferential statistical techniques test hypotheses about differences in populations on the basis of
measurements made on samples of subjects. If reliable differences are found, descriptive statistics
are then used to provide estimations of central tendency, and the like, in the population. Descriptive
statistics used in this way are called parameter estimates.
Use of inferential and descriptive statistics is rarely an either-or proposition. We are usually
interested in both describing and making inferences about a data set. We describe the data. find

S~trategiesl o r random sampling are Jiscussed in rnany sources, including Levy and I-elnenshou i 1990). Rea and Parker
( 1997). and de Vaus (2002).
8 CHAPTER I

reliable clifferences or relationah~p\.and ehtlrnate population ~aluehfor the reliable findings. H o & -
ever. there are more restrictions on inference than there are on description. Many ~~ss~imptions of
multivariate statistical methods are necessary only for inference. If simple description of the sample
is the major goal, many assumptions are relaxed, as discussed in Chapters 5 through 16 and 18
(online).

1.2.4 Orthogonality: Standard and Sequential Analyses

Orthogonality is a'perfect nonassociation between variables. If two variables are orthogonal, know-
ing the value of one variable gives no clue as to the value of the other; the correlation between them
is zero.
Orthogonality is often desirable in statistical applications. For instance, factorial designs for
experiments are orthogonal when two or more IVs are completely crossed with equal sample sizes
in each combination of levels. Except for use of a common error term, tests of hypotheses about main
effects and interactions are independent of each other; the outcome of each test gives no hint as to the
outcome of the others. In orthogonal experimental designs with random assignment of subjects,
manipulation of the levels of the IV, and good cnntro!~, changes i~ va!ue of the DV can be unam-
biguously attributed to various main effects and interactions.
Similarly, in multivariate analyses, there are advantages if sets of IVs or DVs are orthogonal.
If all pairs of IVs in a set are orthogonal, each IV adds, in a simple fashion, to prediction of the DV.
Consider income as a DV with education and occupational prestige as IVs. If education and occupa-
tional prestige are orthogonal, and if 35% of the variability in income may be predicted from educa-
tion and a different 45% is predicted from occupational prestige, then 80% of the variance in income
is predicted from education and occupational prestige together.
Orthogonality can easily be illustrated in Venn diagrams, as shown in Figure 1 . I. Venn dia-
grams represent shared variance (or correlation) as overlapping areas between two (or more) circles.
The total variance for income ic one circle. The section with hori~ontalstripes represents thc part of
income predictable from education, and the section with vertical stripes represents the part pre-
dictable from occupational prestige; the circle for education overlaps the circle for income 35% and
the circle for occupational prestige overlaps 45%. Together. they account for 80% of the variability
in income because education and occupational prestige are orthogonal and do not themselves over-
lap. There are similar advantages if a set of DVs is orthogonal. The overall effect of an IV can be par-
titioned into effects on each DV in an additive fashion.

FIGURE 1.1 Venn diagram for Y

(income), Xi (education), and X z
(occupational prestige).
 Introduction 9

Usually. howe~er.the variables are crJrrelated with each other (nonorthoponal).IV\ !n noneu-
perimental desipns are often correlated naturally: in experimental designs. IV become correlated
when unequal numbers of subjects are measured in different cells of the design. DVs are usually cor-
related because individual differences among subjects tend to be consistent over many attributes.
When variables are correlated, they have shared or overlapping variance. In the example of
Figure 1:2, education and occupational prestige correlate with each other. Although the independent
contribution made by education is still 35% and that by occupational prestige is 4596, their joint con-
tribution to prediction of income is not 80% but rather something smaller due to the overlapping area
shown by the ari-ow in Figure 1.2(a). A major decision for the multivariate analyst is how to handle
the variance that is predictable from more than one variable. Many multivariate techniques have at
least two strategies for handling it; some have more.
In standard analysis, the overlapping variance contributes to the size of summary statistics of
the overall relationship but is not assigned to either variable. Overlapping variance is disregarded in
assessing the contribution of each variable to the solution. Figure 1.2(a) is a Venn diagram of a stan-
dard analysis in which overlapping variance is shown as overlapping areas in circles; the unique con-
tributions of X I and X2 to prediction of Yare shown as horizontal and vertical areas, respectively, and
the total relationship between i.' and the colnbination of X I and X, is those two areas plus the area
with the arrow. If X I is education and X2 is occupational prestige, then in standard analysis, X I is
"credited with" the area marked by the horizontal lines and X2 by the area marked by vertical lines.
Neither of the IVs is assigned the area designated with the arrow. When X I and X2 substantially over-
lap each other, very little horizontal or vertical area may be left for either of them despite the fact that
they are both related to !l They have essentially knocked each other out of the solution.
Sequential analyses differ in that the researcher assigns priority for entry of variables into
equations, and the first one to enter is assigned both unique variance and any overlapping variance it
has with other variables. Lower-priority variables then are assigned on entry their unique and any
remaining overlapping variance. Figure 1.2(b) shows a sequential analysis for the same case as Fig-
ure 1.2(a). where X ! (education) is given priority over X 2 (occupational prestige). The total variance
explained is the same as in Figure 1.2(a), but the relative contributions of X I and X 2 have changed;

Area represents variance

in relationship that contributes
to solution but is assigned to
neither X, nor X,

(a) Standard analysis (b) Sequential analysis i n which

X, is given priority over X,

FIGURE 1.2 Standard (a) and sequential (h) analyses of the relationship
between Z: X,, and X z . Horizontal shading depicts variance assigned to XI.
Vertical shading depicts variance assigned to X L .
10 CHAPTER I

education nou \how\ cl htroriger ~.elation\h~p uith Income than in the st;lndal-d analyhb. wht.re;is the
relation between occupational prestige and income remains the same.
The choice of strategy for dealing with overlapping variance is not trivial. I f variables are cor-
related, the overall relationship remains the same but the apparent importance of variables to the
solution changes depending on whether a standard or a sequential strategy is used. If the multivari-
ate procedures have a reputation for unreliability, it is because solutions change, sometimes dramat-
ically, when different strategies for entry of variables are chosen. However. the strategies also ask
different questions of the data, and it is incumbent on the researcher to determine exactly which
question to ask. We try to make the choices clear in the chapters that follow.

1.3 Linear Combinations of Variables

Multivariate analyses combine variables to do useful work such as predict scores or predict group
membership. The combination that is formed depends on the relationships among the variables and
the goals of analysis, but in most cases. the combination that is formed is a linear combination. A lin-
ear combination is one in which each variable is assigned a weight (e.g.. W i ) ,and then products of
weights and variable scores are summed to predict a score on a combined variable. In Equation I. I,
Y ' (the predicted DV) is predicted by a linear combination of X I and X , (the IVs).

If, for example, Y ' is predicted income, X I is education, and X , is occupational prestige, the
best prediction of income is obtained by weighting education (XI) by W I and occupational prestige
(X,) by W, before summing. No other values of W i and W, produce as good a prediction of income.
Notice that Equation 1.1 includes neither X or X , raised to powers (exponents) nor a product
of X i and X,. This seems to severely restrict multivariate solutions until one realizes that X I could
itself be a of two different variables or a single variable raised to a power. For example, X I
might be education squared. A multivariate solution does not produce exponents or cross-products of
IVs to improve a solution, but the researcher can include Xs that are cross-products of IVs or are IVs
raised to powers. Inclusion of variables raised to powers or cross-products of variables has both the-
oretical and practical implications for the solution. Berry (1993) provides a useful discussion of
many of the issues.
The size of the W values (or some function of them) often reveals a great deal about the rela-
tionship between Dt' and iVs. if, for instance, the W value for some IV is zero, the IV is not needed
i n the best DV-TV relationship. 0: if s=me IV has a large W value, then the i'v' tends to be important
to the relationship. Although complications (to be explained later) prevent interpretation of the mul-
tivariate solution from the sizes of the W values alone, they are nonetheless important in most multi-
variate procedures.
The combination of variables can be considered a supervariable, not directly measured but
worthy of interpretation. The supervariable may represent an underlying dimension that predicts
something or optimizes some relationship. Therefore, the attempt to understand the meaning of the
combination of IVs is worthwhile in many multivariate analyses.
In the search for the best weights to apply in combining variables, computers do not try out all
possible sets of weights. Various algorithms have been developed to compute the weights. Most
algorithms involve n~anipulationof a correlation matrix, a variance-covariance matrix, or a sum-of-
 Introduction 11

squares and cross-products matrlx. Section I .h describes these matrices In \el-? simple term\ ancl
shows their development from a \.cry small data set. Appendix A describes some terms and manipu-
lations appropriate to matrices. In the fourth sections of Chapters 5 through 16 and I8 (online). a
small hypothetical sample of data is analyzed by hand to show how the weights are derived for each
analysis. Though this information is useful for a basic understanding of multivariate statistics, it is
not necessary for applying multivariate techniques fruitfully to your research questions and may,
sadly, be skippe'd by those who are math aversive.

1.4 Number and Nature of Variables to Include

Attention to the number of variables included in analysis is important. A general rule is to get the best
solution with the fewest variables. As more and more variables are included, the solution usually
improves, but only.slightly. Sometimes the improvement does not compensate for the cost in degrees
of freedom of including more variables, so the power of the analyses diminishes.
A second problem is ovetj5tting. With overfitting, the solution is very good, so good in fact,
that it is unlikely to generaiize to a popillatioii. Overfitting occgrs when too many variables are
included in an analysis relative to the sample size. With smaller samples, very few variables can be
analyzed. Generally, a researcher should include only a limited number of uncorrelated variables in
each analysis,6 fewer with smaller samples. We give guidelines for the number of variables that can
be included relative to sample size in the third section of Chapters 5-16 and 18.
Additional considerations for inclusion of variables in a multivariate analysis include cost,
availability, meaning, and theoretical relationships among the variables. Except in analysis of struc-
ture, one usually wants a small number of valid, cheaply obtained, easily available, uncorrelated
variables that assess all the theoretically important dimensions of a research area. Another important
consideration is reliability. How stable is the position of a given score in a distribution of scores when
measured at different times or in different ways'! Unreliable variables degrade an analysis whereas
reliable ones enhance it. A few reliable variables give a more meaningful solution than a large num-
ber of less reliable variables. Indeed, if variables are sufficiently unreliable, the entire solution may
retlect only measurement error. Further considerations for variable selection are mentioned as they
apply to each analysis.

1.5 Statistical Power

A critical issue in designing any study is whether there is adequate power. Power, as you may reca!!,
represents the probability that effects that actually exist have a chance of producing statistical sig-
nificance in your eventual data analysis. For example, do you have a large enough sample size to
show a significant relationship between GRE and GPA if the actual relationship is fairly large? What
if the relationship is fairly small? Is your sample large enough to reveal significant effects of treat-
ment on your DV(s)? Relationships among power and errors of inference are discussed in Chapter 3.
Issues of power are best considered in the planning state of a study when the researcher deter-
mines the required sample size. The researcher estimates the size of the anticipated effect (e.g., an
expected mean difference). the variability expected in assessment of the effect, the desired alpha

"The exceptions are analysis of structure. such as factor analysis. in which numerow correlated variables are measured
12 CHAPTER I

level (ordinarily 0.05). m d the desired power (often ,801.Thehe four estimates are required to deter-
mine necessary sample size. Failure to consider power in the planning stage often results in failure
to find a significant effect (and an unpublishable study). The interested reader may wish to consult
Cohen (1965, 1988). Rossi ( 1990), or Sedlnleier and Giperenzer ( 1989) for more detail.
There is a great deal of software available to help you estimate the power available with various
sample sizes for various statistical techniques, and to help you determine necessary sample size given
a desired level of power (e.g., an 80% probability of achieving a significant result if an effect exists)
and expected sizes of relationships. One of these programs that estimates power for several techniques
is PASS (NCSS, 2002). Many other programs are reviewed (and sometimes available as shareware) on
the Internet. Issues of power relevant to each of the statistical techniques are discussed in chapters cov-
ering those techniques.

1.6 Data Appropriate for Multivariate Statistics

An appropriate data set for multivariate statistical methods consists of values on a number of vari-
ables for each of several subjects or cases. For continuous variables, the values are scores on vari-
ables. For example, if the continuous variable is the GRE (Graduate Record Examination), the values
for the various subjects are scores such as 500,650,420, and so on. For discrete variables, values are
number codes for group membership or treatment. For example, if there are three teaching tech-
niques, students who receive one technique are arbitrarily assigned a "I," those receiving another
technique are assigned a "2," and so on.

1.6.1 The Data Matrix

The data matrix is an organization of scores in which rows (lines) represent subjects and columns
represent variables. An example of a data matrix with six subjects7 and four variables is in Table 1.1.
For example, X I might be type of teaching technique, X , score on the GRE, X 3 GPA, and X4 gender,
with women coded 1 and men coded 2.
Data are entered into a data file with long-term storage accessible by computer in order to
apply computer techniques to them. Each subject starts with a new row (line). Information identify-
ing the subject is typically entered first, followed by the value of each variable for that subject.

TABLE 1.1 A Data Matrix of Hypothetical Scores

Student X~ X2 x3 x4

' ~ o r r n a l l ~of, course. there are many more than six jubjects.
Random documents with unrelated
content Scribd suggests to you:
knowledge of the ethnological conditions under which it is
perpetuated. What public opinion, custom, and long experience have
found to be beneficial and have pronounced not inconsistent with
morality, is very often not a question of ethics, but merely a matter
of expediency. Institutions which nations inhabiting the tropics
defend as necessary could not be adopted without injury by the
sluggish races of the North; and of their propriety, we at a distance
of eight thousand miles are incompetent, not to say prejudiced,
judges. The women who rose to such distinction under the khalifate
were, without exception, members of polygamous households, a
circumstance which would seem to effectually contradict the
prevalent idea that the system of the harem inevitably tends to
intellectual debasement. The standard of morals under the Hispano-
Arab domination was probably much superior to that which now
obtains in the great capitals of Europe. The deplorable condition of
modern society, even among the highly cultivated, where monogamy
nominally exists, is disclosed by the frequency of divorce cases and
the significant revelations of criminal statistics. It demonstrates that
the primitive impulse which among barbarians leads to communal
marriage—the original social state of man—is not only not extinct,
but even generally prevails, although decorously concealed, and,
however repugnant to every principle of morality, must be
recognized as a powerful and retarding element of our boasted
civilization.
The chivalrous courtesy born of intellectual culture and refined
surroundings which distinguished the Spanish Moslems in all the
phases of their social life was, as above stated, eminently
conspicuous in their treatment of females. The latter were, for the
most part, highly educated. Even to-day, in the harems of
Constantinople, it is not unusual to see women fine musicians,
excellent conversationalists, familiar with the principles of art, able to
express themselves fluently in three or four languages. Such
accomplishments are still sufficiently rare to confer distinction upon
their possessors in London, Paris, and New York. Under the khalifs of
the House of Ommeyah, the mental faculties of the sex were
cultivated to a marked degree; no field of literature was closed to
those who aspired to eminence. They were everywhere received
with great respect. They were never insulted in public. They
traversed districts in revolt without molestation. The laws protected
them against the excesses of marital jealousy. If divorced, the wife
was certain of maintenance. It was she who, at marriage, received
the dowry. Public opinion denounced as infamous the husband who
permitted his spouse to labor in order that he might profit by her
earnings. In case of his death she was entitled to a share of his
estate. All things considered, the legal status of woman under the
khalifate appears to advantage when compared with that to which
she is restricted by modern legislation. If polygamy entailed the
unhappiness which foreign prejudice is accustomed to attribute to it,
the practice would long since have been abolished. It is but a natural
result of climatic and physiological conditions, an apparently
indispensable factor in the maintenance of Oriental life.
Slavery in Europe under the Moslems brought with it the
numerous privileges and indulgent treatment enjoined by the
Prophet. The Mohammedan slave was rarely abused or persecuted.
His acceptance of the faith of Islam rendered his manumission easy.
No stigma attached to his condition. He could aspire to the most
noble matrimonial alliance. He was eligible to the most important
political employments. While his master was entitled to exercise
despotic authority over him, the patriarchal customs of the Orient
discouraged all exhibitions of unmerited severity, and designated the
slave rather as a companion than a dependent in the household. It
was contrary to law to put him in chains. His personality was never
sacrificed to the convenience of trade; his classification as a chattel
would have been abhorrent to all Mussulman ideas of justice and
humanity; and in this respect the laws of the Koran are
immeasurably superior to the provisions of Roman and Anglo-Saxon
jurisprudence. An obligation, whose force the lapse of time could
never diminish, was imposed upon the descendants of a freedman to
assist and protect at the risk of their lives all members of the family
which had liberated their ancestor from bondage. The dignity of
human nature was never outraged by the infliction of torture upon
those whom fate had condemned to a state of helpless subjection;
on the contrary, the slave was usually educated by his master; he
became his secretary, his agent, his counsellor; he superintended
the affairs of his family; he executed with diligence and fidelity
important commissions in distant lands.
The cheapness of slaves indicates their abundance; their price
was within reach of the humblest laborer. After the battle of Zallaca,
an ordinary captive could be obtained for a dirhem. Many inmates of
the harems came from the East. Circassian and Georgian girls,
purchased in the markets of Constantinople, were imported into
Spain as early as the ninth century. In Mussulman law a distinction
existed between slaves bought for service and prisoners taken in
battle. The latter shared few of the privileges of the ordinary
bondman, and, strictly speaking, could never be liberated or
ransomed.
The amusements of the Spanish Arabs were derived from the
East. There was nothing in Roman tradition or Visigothic inheritance
which appealed to their imagination like the diversions of the idle
and sensuous races that inhabited the tropics, and which, with other
congenial customs, they had appropriated. They felt but a languid
interest in the chase of ferocious beasts. They shrank with horror
from the gladiatorial contests of the arena and their scenes of blood
and butchery. Exhibitions of strength, where muscular superiority
carried off the palm, were scarcely less distasteful to a people
accustomed to rely for success on fertility of resource and personal
agility. While active exercise was not neglected, those pastimes were
in highest favor which required the least physical exertion. Among
these, the principal one was the game of chess. Of unknown but
high antiquity, it had been brought by Arabic merchants from India.
In that country it had long been used as an instrument of divination,
and, in time of war, the movements of its pieces frequently directed
the evolutions of armies on the march and in battle. A part of the
sacred furniture of every Hindu temple, the board had also a
cabalistic and astrological significance. Long before the appearance
of Mohammed, this game was the solace of the vagrant sheiks of the
Desert and the delight of the wealthy traders of Yemen. It followed
everywhere in the train of the Moslem armies. In Spain it was
universally popular. The chessmen of the khalifs were not inferior in
richness to the other accessories of royal luxury,—the arms, the
plate, the furniture of the palace. Some were made of the precious
metals; others were curiously carved of ivory; most of them were
incrusted with gems. The boards were of ebony and sandalwood
inlaid with gold. In this instance, also, as in many others, the
prohibition of the Koran relating to the representation of animal
forms was disregarded. The Spanish Moslems were passionately
fond of chess. It became one of the favorite diversions of the court;
and it was no unusual occurrence for players to pass the entire day
engrossed by its fascinations and entirely oblivious of their
surroundings. The story, already related, of the prince who pleaded
for time to finish his game after his death-warrant had been read to
him, is an example of the absorbing interest excited by this scientific
pastime in the mind of the Moor. Cards were known to the Arabs
long before the Hegira. Naipe, the Spanish name for them, is from
the Arabic word naib, “viceroy,” whence comes the English “nabob.”
Introduced into Italy by the Saracens, they were at first called The
Game of the Kings. They were not generally used in Europe until the
latter part of the fourteenth century. Backgammon and draughts
were also familiar to the Moors of Spain. The genius of Mohammed
recognized the hidden danger which beset his followers when he
forbade indulgence in all games of chance. To such a temptation the
ardent and romantic nature of the Oriental is peculiarly susceptible.
No information, in this respect, is now obtainable concerning the
Mohammedan population of the Peninsula, but the copious accounts
of the prevalence of other vices under the domination of the emirs
and the khalifs would seem to indicate, from the general silence on
this point, that gaming was not commonly practised.
The feats of jugglers were a source of popular amusement in
mediæval Cordova. These mountebanks were intimately associated
with itinerant minstrels and extemporaneous rhymers, whose coarse
effusions, while they could scarcely be dignified by the name of
poetry, yet often contributed to the diversion of the court, and
whose calling and example produced the troubadour, such an
important agent in the civilization of Europe. The lascivious
contortions of the dancers of ancient Gades, immortalized in the
epigrams of Martial, and which have been transmitted with probably
trifling changes through the Phœnician, Carthaginian, Roman,
Gothic, and Mussulman dominations to the Spanish gypsies of our
day, were constantly exhibited, in all their suggestive indecency,
before the appreciative audiences of Moorish Spain. Nothing can
indicate more positively the general relaxation of manners than the
popularity of such an amusement. Even the indulgent and profligate
spirit of Roman society eyed it with marked disfavor. The poets
lampooned those who patronized or encouraged it. Moralists and
legislators condemned it as a prolific source of corruption.
Mohammed forbade it to his followers as a relic of Paganism and an
incentive to immorality. Under no circumstances did men participate
in it, or, indeed, in any of the terpsichorean exercises practised by
Orientals. The dance, as we understand it, was unknown to the
Moslems. Among them the practice was abandoned to female
professionals, who constituted a caste, who were distinguished by a
peculiar costume, and whose calling was infamous. This prejudice,
descended from a remote antiquity, exists in full force in all Eastern
countries to-day. The degradation of Herodias was far more
reprobated by the Hebrews than her inhumanity. The character of
the bayadere of India, of the ghawazee of Egypt, of the Jewess of
Tunis, of the gypsy of Spain, inheritors of the lewd Phœnician
positions and gestures, is familiar to all travellers.
In the dances of Mohammedan Spain, as in those still practised at
Cairo, the lower limbs were stationary, and all movements were
performed with the body and the arms. Their impropriety generally
consisted rather in their suggestiveness than in any flagrant personal
exposure. Rarely were they performed in a condition of nudity; as a
rule, the form was completely enveloped in graceful folds of silk and
linen. The dancers kept time with castanets, which were originally
small copper cymbals, and every motion was made in perfect
cadence with the music. The extraordinary effect of these exhibitions
upon the imagination, even when represented by women not adepts
in the art, can be understood only by those who have witnessed
them.
The taste for improvisation pervaded the music of the Hispano-
Arabs as it did their poetry. Although to foreign ears it might appear
wholly destitute of measure and harmony, the monotonous
execution of the performer impressed the feelings of his audience to
an extent incomprehensible to nations of northern blood. The
profoundly emotional nature of the Moor, readily susceptible to every
kind of mental excitement and passionately devoted to rhyme, at
one time roused him to frenzy, at another deprived him of
consciousness. No race has ever enjoyed to an equal degree with
the Arabs the faculty of investing fiction with the semblance of truth,
of transforming images created by an inexhaustible fancy into the
realities of life, of giving
—“to airy nothing
A local habitation and a name.”

The music of the khalifate was largely derived from Greek and
Roman sources. Its peculiarities, inherited by the Spaniards, have
imparted a national character to their minstrelsy, as well as
measures unknown to the other nations of Europe. It had nothing in
common with our ideas of harmony. It consisted principally of
monotonous chants, whose time was marked by rude instruments of
percussion; whose melody was partly classic, partly barbaric; and
which disclosed none of that novelty and variety which constitute the
greatest charm of modern music. The Spanish Arabs had no theory,
used no notes, and possessed no means of preserving musical
compositions except by memory and oral tradition. Under these
disadvantages, improvement in one of the most pleasing of sciences
was impossible, and most pieces rarely survived their composers.
The musical instruments of the Spanish Moslems were of many
kinds; there were thirteen different varieties—among them viols,
lutes, dulcimers, harps—made in Seville, which was the most
celebrated seat of their manufacture in the world. The great Ziryab,
who lived at the court of Abd-al-Rahman II., added a fifth string to
the lute, to which, as to the others, the Moors attached a symbolic
significance. The remaining strings, which were of different colors,
represented the supposed four humors of the human body; that of
Ziryab was presumed to represent the soul. The school of music
which he founded at Cordova endured until the last days of the
khalifate, to which is no doubt due the fact that the writers of Spain
on this subject are the most numerous and prolific of any age.
The antipathy with which this science was regarded by
theologians did not prevent it from being the delight of the prince
and an indispensable diversion of the people. Its power was so great
that it early invaded the shrines of the religion that condemned it,
and for centuries the verses of the Koran have been intoned in the
mosques. The story-teller recited his tales to music; the itinerant
buffoon interspersed his coarse but expressive pantomime with
rhyming jests and ribald songs. The Arab notes are harsh, nasal, and
guttural, unpleasant beyond measure to European ears. Their scale
includes seventeen intervals in the octave, and it is said by learned
authority on the subject that the Italian, from which ours is derived,
was originally copied, without alteration, from that of the Arabs of
Sicily and Spain.
The Moors thoroughly understood the almost magical effects
which follow the judicious employment of music. Not only was it
indispensable on all festive occasions, but its notes brought
consolation and comfort to the house of bereavement. It was
considered an important remedial agent in disease. It was used to
correct the distressing condition of insomnia. In the hospitals of
great cities bands were constantly entertained, because it was well
known that harmony of sound promoted convalescence. Its
aphrodisiacal qualities were appreciated and utilized to fully as great
an extent as those of perfumes,—the delight of the Oriental. It was
a favorite maxim of the Mussulman doctors that “to hear music is to
sin against the law; to make music is to sin against religion; to take
pleasure in it is to be guilty of infidelity.” Notwithstanding this, and
the fact that it was anathematized by Mohammed, no people were
more fond of it than the Arabs; and the professional musician,
whose talents had raised him above mediocrity, was sure of
distinguished attention at the court of the khalifs; and, once become
famous, he was the recipient of honors elsewhere reserved for the
descendants and the representatives of royalty. It was not unusual
for a master of his art to receive ten thousand pieces of gold for a
single performance. Instruments of percussion, and especially drums
and tambourines, were most employed by the Spanish Moslems, but
their constructive genius produced radical changes in others; they
improved the guitar, the flute, and the clarionet; they were the
inventors of the mandolin and the organ.
In equestrian sports, which required the highest degree of
adroitness and agility, the Moors of Spain had no superiors. First
among their pastimes of this description was the bull-fight, which
had little in common with the modern spectacle, whose revolting
characteristics are the result of long-continued sanguinary and
brutalizing influences. The performers were all of noble birth; they
were splendidly mounted; their equipments were of the most
sumptuous description. No weapon was allowed them but a short,
heavy javelin, whose point was partly encased in leather. The rules
of the sport required that the animal be killed by a thrust along the
spine in front of the shoulder, to deliver which properly demanded
great skill and almost superhuman strength. If a blow was landed
elsewhere, the knight was compelled to retire from the arena; if his
weapon was broken or lost, he was adjudged to have sustained an
irretrievable disgrace. The intelligence and training of the horse and
the dexterity of the rider were ordinarily sufficient guarantees
against disaster; but the occasional sacrifice of a cavalier reminded
the survivors of the fearful dangers of the encounter. Trained from
early childhood to the use of the horse and the javelin, accustomed
to every manly exercise, adepts in the arts of the tourney and the
chase, the Spanish Moslems found in the bull-fight the climax of
enjoyment, second only to the martial pleasures and excitements of
war. With such an education, it is not strange that they were
recognized as the finest light cavalry in Europe.
Of equal interest, and of even greater magnificence, was the
spectacle presented by the tilt of reeds. In its exhibition and
accessories were displayed the inexhaustible profusion and opulence
of Moorish luxury. The scene was laid in one of the many squares of
the vast Moslem capital. A series of arcades and galleries, supported
by columns of colored marble, brilliant with mosaics and gilded
stuccoes, were crowded with an enthusiastic audience representing
the noblest families of the court and the wealth and fashion of the
principal city of the empire. The Khalif was there, surrounded by his
body-guard, gigantic blacks from the Atlas and the Soudan, with
gem-studded weapons and armor damascened with gold. The
balustrades of the galleries were hung with scarlet, emerald, and
sky-blue velvet. The inmates of the harems, models of the
voluptuous type of Andalusian beauty, unveiled, revealed their
smiling features to the public gaze,—a sight to be witnessed in no
other quarter of the Mohammedan world. Their silken cloaks striped
with every color of the rainbow; their strings of superb jewels,
whose collection was an absorbing passion with every Moorish
woman of rank; the golden belts and bracelets gleaming in the
sunlight; the personal charms of their owners, enhanced to the
utmost by every resource of attire and adornment, presented a
splendid and enchanting picture unsurpassed in either classic or
mediæval times. In the audience, sometimes by courtesy among the
cavaliers in the arena were to be seen Castilian knights exiled for
political reasons, or competitors for distinction in the national sports
of their hereditary foes. The parapets and terraces which
commanded the amphitheatre, the arches of the aqueducts, the
minarets, the trees, even the spurs of the distant sierra, were white
with the robes and turbans of the populace, attracted by the novelty
and magnificence of the scene. The performers, in whom all interest
centred, were worthy of the attention they excited. Twenty-four in
number, they included the flower of the Moorish warriors selected
from two of the principal tribes composing the aristocracy of the
Peninsula. All were clad in flexible coats of mail covered with tunics
of blue or crimson velvet sowed with stars of gold. Their heads were
protected by silken turbans; their waists were encircled by sashes of
the same material; upon the small buckles worn by each horseman
were emblazoned his motto and family crest, from which custom
Christian chivalry borrowed its heraldic devices and its coats of arms.
The horses of one division were white, those of the other black; they
were almost concealed by embroidered housings; the bridles were
enriched with jewels, the bits were of massy gold. A short lance,
whose point was blunted, was the sole arm upon which the cavalier
was permitted to rely for attack or protection; to it were attached
the colors of his mistress,—sometimes represented by a knot of
ribbons, but more frequently by a scarf of silken tissue, upon which
she had traced in golden embroidery the characters of some
amorous legend. In the fantastic devices of Oriental imagery, which
originated in the voluptuous regions where love is an art, each color,
each gesture, even the most prosaic of objects, is invested with
more than a passing significance.
These equestrian diversions of the Spanish Moslems, unlike the
tournaments of the Middle Ages, which were derived from them,
were never polluted by the wanton shedding of blood. They
represented all the exciting phases of battle,—the attack, the mêlée,
the retreat. In the confusing movements of each encounter every
facility was afforded for the exhibition of the highest degree of
strength, activity, and skill. Their object was not the disabling of an
antagonist, but the seizure and retention of the decorations which
adorned both horse and rider; and in the evolutions performed for
the accomplishment of these ends the most daring feats of
horsemanship were exhibited. The course of the rings terminated
the brilliant festival. Among the branches of a tree, planted at one
extremity of the arena, were suspended a number of rings of gold.
One after another the competitors for knightly honors, moving at the
greatest speed, endeavored to bear away these trophies of
adroitness upon the point of the slender reed which served the
purpose of a lance. A magnificent prize, usually a golden vase
enriched with jewels, was awarded the victor, who in turn was
expected to present it to the lady whose colors he had worn in the
contest. The talents of the most famous poets of the khalifate were
exercised in the celebration of these splendid spectacles, wonderful
exhibitions of human dexterity, whose attractiveness was not marred
by suffering, and which revealed to the greatest advantage the
chivalrous sentiments and martial ardor of a refined and polished
race.
Nor were these the only sports which occupied the leisure of the
elegant society of the Moslem court. Its members rarely participated
in the chase. Hawking, introduced by Abd-al-Rahman I., was,
however, a favorite diversion with them; their hawks were the finest
and best trained in Europe, and they constituted an important article
of commerce, especially with Italy, France, and England. In the
extensive gardens of Granada and Palermo were artificial lakes,
where naval spectacles were frequently given upon a much larger
scale than in the amphitheatre of Titus during the palmy days of
Imperial Rome.
Under a government whose beneficent policy provided work for
the industrious and shelter for the helpless, it may well be supposed
that mendicancy was neither an honorable nor a lucrative profession.
The horrible exhibitions of real or simulated deformity which in
Southern Europe now shock the eyes of the traveller were not
tolerated under the Moslem domination. While the bestowal of alms
is a cardinal principle of the Mohammedan religion, its objects, when
worthy, were not permitted to openly solicit the aid of the generous
and sympathetic passer-by. The suffering and the crippled were
carried to hospitals, where every means was applied to effect their
restoration to health; while the impostor, seized by the police,
expiated in prison or under the scourge the penalty of his idleness
and fraud. Conducted on a plan of boundless charity, no factitious
impediments of race or religion interposed to exclude from the public
institutions those unfortunates whose physical afflictions claimed the
indulgence or the generous solicitude of mankind. Hospitals were
open to the worthy applicant, and the Jew, the Christian, even the
idolater, received within their walls the same assiduous care
bestowed upon the most orthodox Moslem.
In all its tendencies the spirit of Moorish civilization was eminently
practical. Even its speculative labors were rather serious than
sportive,—the occasional relaxations of arduous and prolonged
mental effort. Its grand aims were the security of the individual, the
dispensation of impartial justice, the systematic development of the
noblest faculties of the human intellect, the amalgamation of the
heterogeneous constituents of a proud and turbulent society, the
progressive improvement and durable prosperity of a vast and
populous, but constantly disintegrating, empire. In the
accomplishment of these ends, war, while presumed to be an object,
was in reality but an instrument. Public policy required the
occupation of the streams of restless barbarians and needy
adventurers incessantly pouring into the Peninsula. For their
employment in foreign campaigns, the Koranic injunction of
perpetual hostility offered a plausible and convenient excuse. This
practice, while appealing at once to the religious enthusiasm of the
fanatic and the cupidity of the warrior, insured the succession of the
dynasty and the permanence of the throne. Without its aid even the
administration of Al-Mansur, directed by the consummate ability of
that leader, must speedily have fallen. It required semi-annual
campaigns, followed by an unbroken succession of victories, to
restrain the native insubordination of the African immigrants, whose
multitudes, constantly recruited from the innumerable tribes of
Mauritania, constituted not only the bulk of the army, but the
predominant element of the population.
This mingling of races and the resultant prevalence of crosses,
combined with the influence of climate and the stimulants of military
and commercial activity, will readily account for the versatility of the
Hispano-Arab mind, which was among its most prominent
characteristics. No greater contrast in comparative ethnology can be
drawn than that presented by the precarious and barbaric existence
of the Desert and the polished and highly cultivated life of the
Western Khalifate in its most glorious days. And yet but a
comparatively insignificant period of time separates the vagrant
Bedouin, whose favorite occupation was the plunder of his neighbors
and who resented the interference of even his acknowledged
chieftain as an infringement of his liberties, and the Spanish Arab,
whose despotic government insured the enjoyment of personal
freedom and public tranquillity, where intelligence, order, prosperity,
took the place of insubordination and discord, and the prestige of
foreign conquest and the blessings of civilization travelled in parallel
lines and side by side. To the development of that civilization every
people became tributary, coincident with its subjection to the
Moslem arms. In the character of the conqueror was revealed a
spirit of acquisition in no wise inferior to its inventive faculty, and
which at once appropriated, and often improved, all that was useful
in the systems of others while forming and developing new ideas of
its own.
It is with mingled sentiments of admiration and regret that we
contemplate the phenomenal rise, the dazzling splendor, the rapid
fall, of the Moslem empire in Spain. The material relics which remain
to tell the story of its architectural grandeur, of the munificence of its
sovereigns, of the acquirements of its scholars, of the skill of its
artisans, are few and widely scattered. The destruction has been
most complete. The supremacy of Christian ideas and Castilian
customs, enforced by diligent persecution, was in all instances
necessary before the intellectual aspirations fostered during nine
generations of august and learned princes could be subordinated to
the sacerdotal ignorance and military ferocity of the age. Some
edifices defaced by malice or neglect, their apartments so altered by
barbarous innovators that their original plans and the purposes of
their construction can often no longer be traced or even
conjectured; their delicate ornamentation concealed by many
successive layers of lime and plaster; their precincts abandoned to
the vilest uses; a meagre collection of manuscripts, whose
characters are half obliterated by moisture and rough usage; an
occasional trophy rusting in the solitude of the museum, are all the
tangible evidences extant of a monarchy once the marvel of Europe.
It is elsewhere that we must look for the proofs of its greatness and
the trophies of its glory. Its salutary influence in modifying the
debased instincts and savage manners of mediæval society is no
longer questioned. The enduring impulse it imparted to philosophical
investigation, its prosecution of the exact sciences, the consideration
in which it held intellectual ability, the honors with which it rewarded
proficiency in literature, transmitted through many generations, have
placed their seal upon the civilization of the twentieth century. The
obligations we are under to the Spanish Arabs cannot be too
frequently nor too generally acknowledged; and in ascribing the
origin of our progress to the nation whose genius was its inspiring
spirit, we are only offering a just and well deserved tribute of
gratitude.
It was said by Seneca, “Wherever the Roman conquers, he
inhabits.” It might, with almost equal truth, be asserted of the Arab
that, wherever his religion and his language are once established,
there they will forever prevail. The countries originally subdued by
the lieutenants of the Prophet are still Mohammedan. The idiom of
Mecca is still spoken from the eastern shore of the Atlantic to the
China Sea. Nor does Islam seem to have lost its power of expansion.
Its progress has never been arrested. It has penetrated to Central
Russia,—in that empire its votaries number eleven millions. It is the
faith of hundreds of thousands of Negroes at the equator. In Europe
there are seven million Mussulmans, in India fifty-three million, in
China twenty-two million. The people of Sicily and Spain alone of the
great colonies founded by the Moslem—the seat of his most highly
developed civilization, the home of races equally accomplished in
learning, advanced in arts, illustrious in arms—were compelled to go
into exile or renounce their faith and abandon their language. In
neither of these countries have the discoveries, the inventions, and
the experience of six centuries, which have long been the common
property of all nations, exerted any appreciable effect in repairing
the awful damage consequent on Moorish expulsion.
The propagators of a form of religion which relies for its success
upon the extermination of all who refuse assent to its dogmas have
certainly little faith in the truth or the celestial inspiration of the
maxims which they deem it necessary to resort to force to inculcate.
During the ascendency of the papal power no one within its reach
could publicly profess heretical doctrines and live. Under the
Ommeyades and the Aghlabites both the misbeliever and the infidel
were safe on the payment of tribute. The occasional outbursts of
Moslem fanaticism were directed against literature; the spirit of
Christian persecution—a spirit sadly at variance with that evinced by
the gentleness and meekness of its Divine Founder—raged fiercely
against both literature and humanity. Amru and Al-Mansur burned
books. Innocent III. and Calvin tortured men.
The Assyrian, Carthaginian, Roman, and Hispano-Arab empires
lasted each about eight hundred years. Of two of these the memory
alone survives. A number of defaced monuments, a fragmentary
literature, preserve the traditions of the third. The genius of the
Moslem, superior to those of all his predecessors, has perpetuated
itself in the scientific inspiration and progressive energy of every
succeeding age. Remarkable for its unparalleled success, while
hampered by tremendous obstacles,—war, sedition, disorder,
barbarian supremacy,—it is instructive to reflect what it might have
accomplished under the most favorable auspices, when at the height
of their prosperity the Moors of Europe controlled the Mediterranean.
The latter occupied and colonized in turn the important posts of
Sardinia, Corsica, Cyprus, Malta. Their revenues were tenfold greater
than those of any contemporaneous state. The inexhaustible
population of Africa could be constantly drawn on for hundreds of
thousands of soldiers, whose abstemious lives, blind fanaticism, and
reckless bravery made them most formidable adversaries. The fleets
of the Sicilian emirs threatened the coast of Asia Minor. The Arab
governors of Spain established permanent outposts as far as Lyons.
The pirates of Fraxinet fortified and held for many years the passes
of the Alps. The tracks of the Saracen armies marching northward
from Calabria and southward from Provence and Switzerland
overlapped on the plains of Lombardy. Such opportunities for
conquest have rarely been enjoyed or neglected by any military
power. Civil discord and tribal jealousy were all that prevented
Europe from being Mohammedanized. In the polity of the Arabs,
wherever domiciled, the traditions of the Desert invariably prevailed.
The organization of the state was modelled after those of the family
and the tribe. No allowance was made for the changed conditions
resulting from the extension of dominion and the increase of
knowledge. Under such circumstances there could be no cohesion
among the parts of the constantly tottering fabric of Moslem power,
which, in fact, was undermined from the very beginning. From this
instability, the Western Khalifate has been, with some truth,
compared to a Bedouin encampment. The defects of an anomalous
constitution were aggravated by intestine conflict. Factional hostility
in Arabian Spain was always more pronounced and bitter than
hatred of the Christian foe.
A great victory and a few unimportant skirmishes gained for the
Moslems in less than two years control of the richest kingdom in
Europe. To reconquer it required eight centuries and more than five
thousand battles. The followers of Pelayus, when the long struggle
for Christian supremacy began, were but thirty in number. The host
mustered by Ferdinand and Isabella before Granada amounted to
nearly a hundred thousand men. The religious character which
invested the Reconquest, and from which its prosecution eventually
became inseparable, has stamped itself indelibly and ruinously upon
the Spanish people. The cost of the triumph was incalculable. It
impoverished forever great provinces. It drenched the soil of the
entire Peninsula with blood. A single campaign often destroyed an
army. The casualties of a single siege sometimes swept away
numbers equal to the inhabitants of a populous city. At Baza alone,
in the short space of seven months, twenty-one thousand Castilians
perished.
The almost universal disbelief in Moorish civilization is hardly less
remarkable than its creation and progress. Sectarian prejudice,
ignorance of Arabic, and a fixed determination to acknowledge no
obligation to infidels have concurred to establish and confirm the
popular opinion. To this end the Church has always lent its powerful,
often omnipotent, aid. Yet, in spite of systematic suppression of
facts and long-continued misrepresentations, it cannot now be
denied that no race effected so much for all that concerns the
practical welfare of mankind as the Spanish Mohammedans; that no
race of kings has deserved so large a measure of fame as that which
traced its lineage to Abd-al-Rahman I.
Such was the civilization which the Spanish and Sicilian Arabs
bequeathed to Europe. Their conquests and their influence, their
progress in the arts of peace, their industrial and economical
inventions, the precocity of their mental development, the
perpetuation of their advanced ideas under the most discouraging
conditions which can be conceived, present an example without
parallel in the history of nations. Their origin had nothing in common
with that of any European people. Their religion was avowedly
inimical to the one which was professed from the Mediterranean
coast to the verge of the Arctic Circle. Their political and domestic
institutions were abhorrent to the feelings of their neighbors, their
allies, their enemies. From the hour when Tarik landed at Gibraltar to
that when Boabdil surrendered the keys of the Alhambra was a
period of constant and relentless hostility. Such circumstances as
these are not ordinarily propitious to the material or intellectual
advancement of mankind.
In the face of such formidable obstacles a mighty empire was
founded. The very causes which seemed liable to seriously affect its
integrity and permanence in reality increased its strength. Its military
power became a standing menace to every state of Christendom. Its
fleets of armed galleys dominated the seas. The Saracens of Sicily
sacked the suburbs of Rome and insulted the sacred majesty of the
Holy Father in the Vatican. In every trade-centre of the East and
West, in the streets of Canton and Delhi, in the bazaars of
Damascus, along the crowded quays of Alexandria, beside the
scattered wells of the Sahara, at the great fairs of Sweden, Germany,
and Russia, in the splendid markets of Constantinople, the Moorish
merchants and Hebrew brokers of Spain outstripped all commercial
competitors in the amounts of their purchases and the shrewdness
of their bargains. The wealth which resulted from this vast system of
trade was almost inconceivable. In addition, the agricultural and
mineral resources of the country, great in themselves, were
developed beyond all precedent. The treasures thus amassed were
expended in public works, whose neglected ruins amaze the
traveller; in the promotion of educational advantages that modern
experience and energy have never been able to surpass; in the
collection of immense libraries; in the maintenance of a court with
whose magnificence the traditional luxury of the Byzantine princes
was not worthy of comparison; in the celebration of a worship
whose furniture and appointments transcended, in richness and
beauty, the vaunted pomp and semi-barbaric ceremonial of pontifical
Rome.
It is both popular and fashionable to ascribe to the influence of
the Crusades the awakening of the spirit of progress which
ultimately led to the revival of letters and to the political and social
regeneration of Europe. But the Crusades were only, in an indirect
and secondary manner, a factor of civilization. On the other hand,
their general tendency was signally destructive. Their track has been
compared to that left by a swarm of locusts. Many works of classic
genius perished in the sack of Constantinople. The Moslem library of
Tripoli, which contained two hundred thousand volumes, was burned
when that city was taken by the soldiers of the Cross. It is a well-
established fact that few of the latter were actuated by religious
motives. Their crimes cast discredit upon their cause and secured
the eternal contempt of the Oriental; for even the name of
Christianity was unworthily degraded by such vile associations. The
results produced upon Europe by these expeditions, instead of being
humanizing, were most disastrous. Whole districts were
depopulated. The hereditary estates of the nobility were transferred
to the Church, whose ministers alone possessed the means of
purchase, and who, through promoting the insane spirit of
fanaticism by which they subsequently profited, secured a double
measure of consequence and power. The Papacy soon controlled the
wealth of Christendom, and its irresponsible authority increased in
proportion to its influence. With despotism came tyranny, with
tyranny persecution. The principle of forcing the acceptance of
religious dogmas upon armed enemies was extended to the
conviction of recalcitrant sectaries by torture. The atrocities of
religious conflict, the war of the Albigenses, the unspeakable horrors
of the Inquisition, the massacre of St. Bartholomew, are largely
attributable to the sanguinary tastes engendered by the Crusades. In
other respects, as well, their influence was highly detrimental to
humanity. They introduced vices hitherto confined to the East, and
which are to this day blots upon the society of the great European
capitals. They filled Europe with leprosy or with an affection similar
to it, from which eminent medical authorities have deduced the
origin of the most obstinate and loathsome of contagious diseases.
They introduced the plague, one visitation of which swept away
thirteen million persons. The rupture of family ties occasioned by the
absence of such multitudes fostered every form of licentiousness. In
some provinces vast tracts of fertile soil, soon overgrown with
brushwood, relapsed into primeval wildness. In others, deprived of
the means of preserving order, the country became a prey to
outlaws. While tens of thousands of armed fanatics were fighting for
the Christian cause in Syria, the barbarians of Northern Europe were
worshipping idols and serpents and offering human sacrifices.
The Crusades, however, were not wholly an unmixed evil. They
increased the power of the clergy, but they exterminated a large part
of the most worthless elements of society. It has been estimated
that six million persons perished in these expeditions. They made
the Papacy autocratic; but, by destroying feudalism through the
alienation of the estates of the barons, they greatly improved the
condition of the serf. The necessity for treating victims of the
horrible maladies contracted in Palestine led to the foundation of the
first hospitals in Christendom. They directed the attention of scholars
to the study of works in Arabic, a language hitherto unknown
outside of Mussulman countries. It was in 1142 that Peter the
Venerable, Abbot of Cluny, went to Toledo and made a translation of
the Koran into Latin, in order that he might demonstrate the falsity
of the doctrines of Islamism. Had these successive deluges of
fanatics been poured upon the Spanish Peninsula instead of upon
the Holy Land, not the slightest trace of Moslem learning and
civilization could have survived their attack.
The benefits arising from the Crusades were far from sufficient to
counterbalance their injurious effects. They gave, however, a great
impetus to commerce, especially through the enterprise of the
Italian republics. They awakened a taste for luxuries which had been
hitherto unknown, even to royalty. They stimulated manufactures,
particularly those connected with the ornamental arts of glass,
wood, ivory, and metals. In one respect, their influence promoted
immensely the cause of civilization. Familiarity with Moslem valor,
politeness, and culture removed the prejudices maintained through
centuries of priestcraft and ignorance by the benighted nations of
Europe. Returning pilgrims and adventurers brought back from the
Holy Land tales of magnificent cities, of incredible treasures, of
deeds of heroism and chivalry, which had no counterparts in any
state of Christendom. Accounts of these marvels awakened not only
a desire to imitate them, but aroused an involuntary admiration for
the superiority of their authors. At the time of the first Crusade, in
the closing years of the eleventh century, Moorish civilization in the
Peninsula had attained its highest perfection. While its influence had
long been imperceptibly exerted upon the populations of France and
Italy, deep-seated hatred of the followers of Mohammed had
retarded the general diffusion of its benefits. In consequence of the
repeated expeditions to Palestine, an increased demand for the
manufactures and the agricultural products of Moorish Spain was
created. Its language, its improvisations, its literature, soon became
familiar to Europe. Even its sports were borrowed, and the graceful
courses of the arena, adapted to the rude and ferocious tastes of
baronial society, became the most popular of mediæval diversions.
The chivalric sentiments inseparable from knightly exercises
contributed to social refinement and to the exaltation of woman. The
troubadour carried everywhere the amatory songs which had long
enchanted the polished society of Andalusia. The coarseness and
asperity of feudal manners were softened, and a marked
improvement characterized every form of official and domestic
intercourse. It is beyond the Pyrenees, and not to the Orient, that
the historian must look for the origin of modern civilization.
In rapidity of conquest, in extent of dominion, in successful
propagation of religious belief, in ability to profit by the resources of
Nature, in profundity of knowledge and versatility of intellect, no
people have ever approached the Arabs. Their conquests were
secured, and their government made permanent, by that peculiar
provision of their civil polity which, appealing to the strongest of
human passions and sanctioned by the injunctions of their Prophet,
permitted the appropriation of the women of vanquished nations.
Their commerce, to which in a land destitute of agricultural
resources they were impelled by necessity, developed their trading
propensities, and by association from a remote age with their
enterprising neighbors, the Phœnicians, familiarized them with the
men of all races and the products of all countries; enlarged their
faculties; sharpened their intellects; and made them capable of
becoming, in after times, the conquerors and the lawgivers of the
world. Prodigious energy and aggressiveness were their leading
characteristics. These traits were intensified by various, sometimes
by unworthy, motives,—by the love of pleasure, the thirst of avarice,
the fire of ambition,—as well as by the precepts and promises of a
religion congenial to their tastes, their habits, and their excessively
romantic and adventurous nature. Of all the dynasties established by
the Successors of the Prophet, that of the Ommeyades of Spain is
indisputably entitled to the most exalted rank.
The foundation of that dynasty marks a great epoch in the history
of Europe. Of its noble deeds, in both war and peace, every
individual of Moslem faith or Arab lineage may well be proud; proud
of its long line of illustrious princes; proud of its conquests; proud of
its civilization, which surpassed the splendors of Imperial Rome, and
whose arts modern science has found it impossible to successfully
imitate; proud of its unequalled agricultural prosperity; proud of the
exquisite beauty of its edifices, still pre-eminently attractive even in
their decay; proud of its mighty capital; proud of its academical
system, with its perfect organization, its colleges, its lyceums, its
libraries; proud of the vast attainments of its scholars, its surgeons,
its chemists, its botanists, its astronomers, its mathematicians;
proud of the theories of its philosophers, which for a thousand years,
amidst the incessant fluctuations of human opinion and the infinite
variations of religious belief, have retained their original form, and
are accepted as correct by the most enlightened thinkers of the
present age. The destruction of this wonderful empire was an event
of more than national significance; it was a misfortune to be
deplored by every lover of learning for all coming time. For evil was
the day for human progress when from his battlemented walls the
Moor looked down upon the signing of a truce craftily devised for the
betrayal of his kindred; evil was the day when upon the red towers
of the Alhambra, decorated by the emirs with profuse and
unexampled magnificence, and which for seven centuries had been
the stronghold of Moslem power, the home of Moslem art, were
raised the victorious banners of the Spanish monarchy, suggestive, it
is true, of incredible achievement, of undaunted valor, of heroic self-
sacrifice, of imperishable renown, yet at the same time harbingers of
an endless train of national calamities which, like avenging and
relentless furies, stalked unseen in the wake of the exultant
conqueror.
 INDEX

Abbeys of France and England, their extent and wealth, iii. 351.
Abd-al-Aziz, first Emir of Spain, i. 267;
marries widow of Roderick, 269;
is assassinated, 271.
Abdallah gains crown by treachery, i. 535;
character of, 561.
Abd-al-Melik, emir, i. 306;
is impaled, 317.
Abd-al-Mumen, ruler of the Almohades, ii. 259;
conquers Spain, 287.
Abd-al-Rahman-al-Ghafeki conducts retreat, i. 277;
becomes emir and is deposed, 287;
is again raised to that office, 292;
attempts conquest of France, 295;
defeated and killed at Poitiers, 305.
Abd-al-Rahman I., his early career, i. 384, 385;
escapes to the Desert, 385;
lands in Spain, 389;
conquers the Peninsula, 393, 394;
his death and character, 408–411.
Abd-al-Rahman II., ability of, i. 475;
receives embassies from the East, 478, 479;
builds a navy, 491;
his death, 494.
Abd-al-Rahman III. ascends the throne, i. 563;
his noble qualities, 564;
subdues the rebels, 567;
death of, 596;
 his fame, 597;
domestic policy of, 605;
his patronage of letters, 631.
Abd-al-Rahman IV., ii. 85;
his independence, 93.
Abd-al-Rahman V., ii. 99.
Abu-Abdallah, the Mahdi, ii. 249;
character of, 250.
Abu-Bekr, chief of Almoravides, ii. 194;
deposed, 196.
Abul-Hassan, Sultan of Fez, invades Spain, ii. 476.
Abul-Kasim-Mohammed, Kadi of Seville, great power of, ii. 116;
death of, 140.
Abu-Said betrayed by Pedro el Cruel, ii. 492.
Africa, innate barbarism of, ii. 88.
Agriculture, system of the Arabs, its perfection, iii. 599–601.
Aguilar, Alonso de, death of, iii. 250.
Ajarquia, rout of, ii. 562, 563.
Alarcos, battle of, ii. 311.
Albigenses, rise and doctrines of, iii. 90;
crusade against, 95, 96.
Alfonso I., King of Aragon, raid of, ii. 263, 264;
defeated and killed at Fraga, 269.
Alfonso I., King of the Asturias, i. 357;
his expeditions, 359;
his death, 361.
Alfonso III., exploits of, i. 532.
Alfonso VI., reforms of, ii. 162;
prowess of, 183;
enters Toledo, 185.
Alfonso VIII. wins battle of Las Navas, ii. 331.
Alfonso X., great talents of, ii. 441;
literary works, 443;
his death, 444.
Alfonso XI., death of, before Gibraltar, ii. 483.
Al-Hakem I. ascends the throne, i. 440;
 defeats his uncles, 443, 444;
quells rebellion of southern suburb, 466;
his sufferings and death, 474.
Al-Hakem II., accession of, i. 636, 637;
character of, 668;
his love of learning, 670;
his great library, 672;
his erudition, 673;
attempts at reform, 676;
public works, 677.
Alhambra, origin of, i. 547;
magnificence, ii. 525;
gardens, 529.
Alhandega, battle of, i. 588.
Al-Haytham-Ibn-Obeyd appointed emir by the Khalif, i. 290.
Al-Horr appointed emir by the Viceroy of Africa, i. 272.
Ali, ruler of Spain, ii. 87;
his severe measures, 89, 90;
his death, 93.
Al-Maghreb, its extent and fertility, i. 134;
invasion of, by Abdallah, 138;
by Ibn Hajij, 141;
is conquered by Okbah, 143;
is invaded by Hassan, 145;
is finally subjugated by Musa, 162.
Al-Mansur—see Ibn-abi-Amir.
Almohades, rise of, ii. 255.
Al-Mondhir, character of, i. 533;
is poisoned, 535.
Almoravides, origin of, ii. 191;
they conquer Africa, 194;
their immense empire, 239.
Al-Morthada, ii. 91.
Al-Nazer, King of Granada, ii. 454.
Al-Samh, Emir, i. 273;
invades France, 276;
 is killed, 277.
Al-Zagal defeats Christians, ii. 563;
becomes king, 591;
abdication of, 664.
Al-Zarkal, clepsydra of, ii. 164;
quadrant of, iii. 435;
suggests elliptical orbit, 477.
Amulets of Arabs, i. 36.
Anbasah-Ibn-Sohim succeeds Abd-al-Rahman, i. 287;
his severity, 288;
invades Septimania, 290.
Arabia, topography of, i. 1;
dearth of history, 4;
visited by Phœnicians, 5;
its great wealth, 7;
exemption from foreign influence, 10.
Arabs, their prominence in antiquity, i. 16;
their energy, 16;
predatory instincts predominant, 16;
influence of the sheik, 19;
difference from other pastoral nations, 19;
blood revenge, 25;
habits of life, 27, 28;
treatment of woman, 28;
idolatry, 30;
relationship with Jews, 32;
trade of, 39;
wonderful career of the race, 54;
rebel after death of Mohammed, 128.
Architecture under the Moors of Spain, iii. 537–540.
Art, absence of, in Arabia, iii. 535.
Asturias, foundation of the kingdom of, i. 341.
Aurora, sultana, intrigues with Ibn-abi-Amir, i. 691;
opposes the latter, 735.
Averroes, iii. 473–475.
Ayub-Ibn-Habib, provisional emir, i. 271;
 is deposed by the Khalif, 272.

Badis, King of Granada, ii. 134.

Balj-Ibn-Beshr besieged in Ceuta, i. 314;
relieved by Abd-al-Melik, 316;
seizes authority, 317.
Barcelona taken by the Franks, i. 450.
Baths, iii. 643;
luxury of, 644.
Baza, siege of, ii. 651;
capitulation of, 663.
Bedouins, life and character of, i. 17.
Beni-Khaldun, clan of, i. 552.
Berbers, origin and characteristics of, i. 136;
language and government, 137, 138;
oppressed by Arabs, 313, 325.
Bermudo, King of Leon, renders homage to Al-Mansur, i. 727.
Bernhart, count of Barcelona, killed, i. 492.
Béziers, destruction of, iii. 98.
Biscay, its ruggedness and severe climate, i. 338.
Black Stone of Kaaba, i. 35.
Boabdil taken prisoner at Lucena, ii. 568;
released, 572;
his worthless character, 594.
Botany of Spanish Moslems, iii. 486, 487.
Byzantine Empire, condition of, after barbarian conquest, i. 70;
its society and its policy, iii. 370–372;
degradation of all classes, 381, 382.

Calligraphy, skill in, iii. 590.

Carmona taken by Arabs, i. 235;
its siege raised by Abd-al-Rahman I., 400.
Carthage, the ancient city, its origin and splendor, i. 147, 148;
trade, 148;
religion, 151;
buildings, 152;
the Megara, 152;
the Roman city, its arts, its learning, and its vices, 153;
stormed by Hassan, 154.
Castrogiovanni, first attack on, ii. 18;
surprised by Moslems, 29.
Chakya, the impostor, rebels against Abd-al-Rahman I., i. 401;
his defeat and death, 402.
Charlemagne invades Spain, i. 405.
Charles Martel, character of, i. 302;
hated by the clergy, 303;
invades Provence, 309.
Chemistry, its great progress in the Peninsula, iii. 490–492.
Chess, game of, introduced by the Arabs, iii. 661, 662.
Christian tributaries of the Moors, iii. 183;
their tribute and their privileges, 184, 185;
disabilities of, 186, 189;
persecution of, by the khalifs, 204–206.
Christianity made no progress in Arabia, i. 41.
Church, condition of, before Mohammed, i. 66.
Cid, rise of, ii. 160;
character and career of, 220;
valor of, 224;
duplicity of, 226;
takes Valencia, 235;
horrors of the siege, 236;
his death, 237.
Civil organization of the Arabs, iii. 638.
Clergy, influence of, among the Visigoths, i. 175;
their luxury, 194, 211;
increasing power of, ii. 420;
wealth of, 422.
Commerce, its great extent under the Moors, iii. 616–619.
Cordova, beauty and wealth of, under the Ommeyades, i. 618,
619;
suburbs of, 622;
taken by Ferdinand III., ii. 366.
Count Julian resents outrage on his daughter, i. 221;
enters Spain, 224;
retires to Ceuta, 259.
Covadonga, battle of, i. 350.

Damascus, beauty and wealth of city, i. 370–372.

Dances derived from the Orient, iii. 663.
De Hauteville, House of, ii. 54.

Edrisi, geography of, ii. 71.

Egilona, queen of Roderick, captured at Merida, i. 246;
marries Abd-al-Aziz, 269.
Egiza, his tyranny, i. 216.
Egypt, effect of its civilization on the Arabs, i. 132.
Elvira, foundation and wars of, i. 542–549;
surrenders to Abd-al-Rahman III., 567.
Emirate, disorders of, i. 322.
Equestrian sports, their magnificence, iii. 491, 492.
Ervigius, reign and death of, i. 214.

Fatimites of Africa, i. 580;

remove capital to Egypt, 646.
Favila, King of Asturias, i. 356.
Fayic and Djaudar, eunuchs, conspiracy of, i. 697.
Ferdinand Gonzalez, Count of Castile, his character, i. 589;
power and exploits of, 603, 604.
Ferdinand III., character of, ii. 416.
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookfinal.com