Multiple Regression and Beyond An Introduction

to Multiple Regression and Structural Equation

Modeling 2nd Edition Timothy Keith pdf download

https://ebookname.com/product/multiple-regression-and-beyond-an-
introduction-to-multiple-regression-and-structural-equation-
modeling-2nd-edition-timothy-keith/

Get Instant Ebook Downloads – Browse at https://ebookname.com

Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Interaction Effects in Multiple Regression 2nd Edition

James Jaccard

https://ebookname.com/product/interaction-effects-in-multiple-
regression-2nd-edition-james-jaccard/

Introduction to Regression Modeling 1st Edition Bovas

Abraham

https://ebookname.com/product/introduction-to-regression-
modeling-1st-edition-bovas-abraham/

Introduction to Mixed Modelling Beyond Regression and

Analysis of Variance 2nd Edition N. W. Galwey

https://ebookname.com/product/introduction-to-mixed-modelling-
beyond-regression-and-analysis-of-variance-2nd-edition-n-w-
galwey/

intelligent engineering systems through artificial

neural networks volume 19 Computational Intelligence in
Architecting Engineering Systems 1st Edition Cihan H.
Dagli
https://ebookname.com/product/intelligent-engineering-systems-
through-artificial-neural-networks-volume-19-computational-
intelligence-in-architecting-engineering-systems-1st-edition-
The Athenian Experiment Building an Imagined Political
Community in Ancient Attica 508 490 B C Greg Anderson

https://ebookname.com/product/the-athenian-experiment-building-
an-imagined-political-community-in-ancient-attica-508-490-b-c-
greg-anderson/

Lecture Notes Respiratory Medicine 8th Edition Stephen

J. Bourke

https://ebookname.com/product/lecture-notes-respiratory-
medicine-8th-edition-stephen-j-bourke/

Foreign Aid Reform 1st Edition Finn C. Hudson

https://ebookname.com/product/foreign-aid-reform-1st-edition-
finn-c-hudson/

Trojan Odyssey A Dirk Pitt Novel 1ST Edition Clive

Cussler

https://ebookname.com/product/trojan-odyssey-a-dirk-pitt-
novel-1st-edition-clive-cussler/

Modular Cognitive Behavioral Therapy for Childhood

Anxiety Disorders Guides to Individualized Evidence
Based Treatment 1st Edition Bruce F. Chorpita

https://ebookname.com/product/modular-cognitive-behavioral-
therapy-for-childhood-anxiety-disorders-guides-to-individualized-
evidence-based-treatment-1st-edition-bruce-f-chorpita/
Encyclopedia of special education 3 vols 3rd ed Edition
C. Reynolds

https://ebookname.com/product/encyclopedia-of-special-
education-3-vols-3rd-ed-edition-c-reynolds/
 Multiple Regression
and Beyond

Multiple Regression and Beyond offers a conceptually oriented introduction to multiple

regression (MR) analysis and structural equation modeling (SEM), along with analyses that
flow naturally from those methods. By focusing on the concepts and purposes of MR and
related methods, rather than the derivation and calculation of formulae, this book introduces
material to students more clearly, and in a less threatening way. In addition to illuminating
content necessary for coursework, the accessibility of this approach means students are more
likely to be able to conduct research using MR or SEM—and more likely to use the methods
wisely.

• Covers both MR and SEM, while explaining their relevance to one another
• Also includes path analysis, confirmatory factor analysis, and latent growth modeling
• Figures and tables throughout provide examples and illustrate key concepts and
techniques

Timothy Z. Keith is Professor and Program Director of School Psychology at University of

Texas, Austin.
This page intentionally left blank
Multiple Regression
and Beyond
An Introduction to
Multiple Regression
and Structural
Equation Modeling
2nd Edition

Timothy Z. Keith
Second edition published 2015
by Routledge
711 Third Avenue, New York, NY 10017

and by Routledge
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN

Routledge is an imprint of the Taylor & Francis Group, an informa business

© 2015 Taylor & Francis

The right of Timothy Z. Keith to be identified as author of this work has been asserted by
him in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.

All rights reserved. No part of this book may be reprinted or reproduced or utilized in any
form or by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying and recording, or in any information storage or retrieval system,
without permission in writing from the publishers.

Trademark Notice: Product or corporate names may be trademarks or registered

trademarks, and are used only for identification and explanation without intent to infringe.

First edition published by Pearson Education, Inc. 2006

Library of Congress Cataloging-in-Publication Data

Library of Congress Control Number: 2014956124

ISBN: 978-1-138-81194-2 (hbk)

ISBN: 978-1-138-81195-9 (pbk)
ISBN: 978-1-315-74909-9 (ebk)

Typeset in Minion
by Apex CoVantage, LLC
 Contents

Preface vii

Acknowledgments xi

Part I Multiple Regression 1

1 Introduction: Simple (Bivariate) Regression 3

2 Multiple Regression: Introduction 26

3 Multiple Regression: More Detail 44

4 Three and More Independent Variables and Related Issues 57

5 Three Types of Multiple Regression 77

6 Analysis of Categorical Variables 108

7 Categorical and Continuous Variables 129

8 Continuous Variables: Interactions and Curves 161

9 Multiple Regression: Summary, Assumptions, Diagnostics,

Power, and Problems 182

10 Related Methods: Logistic Regression and Multilevel Modeling 213

Part II Beyond Multiple Regression: Structural Equation Modeling 241

11 Path Modeling: Structural Equation Modeling

With Measured Variables 243

v
vi • CONTENTS

12 Path Analysis: Dangers and Assumptions 267

13 Analyzing Path Models Using SEM Programs 282

14 Error: The Scourge of Research 318

15 Confirmatory Factor Analysis I 332

16 Putting It All Together: Introduction to Latent Variable SEM 371

17 Latent Variable Models: More Advanced Topics 391

18 Latent Means in SEM 424

19 Confirmatory Factor Analysis II: Invariance and Latent Means 455

20 Latent Growth Models 493

21 Summary: Path Analysis, CFA, SEM, and Latent Growth Models 514

Appendices
Appendix A: Data Files 537

Appendix B: Review of Basic Statistics Concepts 539

Appendix C: Partial and Semipartial Correlation 557

Appendix D: Symbols Used in This Book 565

Appendix E: Useful Formulae 567

References 569

Author Index 579

Subject Index 583

 Preface

Multiple Regression and Beyond is designed to provide a conceptually oriented introduction to

multiple regression along with more complex methods that flow naturally from multiple regres-
sion: path analysis, confirmatory factor analysis, and structural equation modeling. Multiple
regression (MR) and related methods have become indispensable tools for modern social science
researchers. MR closely implements the general linear model and thus subsumes methods, such
as analysis of variance (ANOVA), that have traditionally been more commonplace in psycho-
logical and educational research. Regression is especially appropriate for the analysis of nonex-
perimental research, and with the use of dummy variables and modern computer packages, it is
often more appropriate or easier to use MR to analyze the results of complex quasi-experimental
or even experimental research. Extensions of multiple regression—particularly structural equa-
tion modeling (SEM)—partially obviate threats due to the unreliability of the variables used in
research and allow the modeling of complex relations among variables. A quick perusal of the full
range of social science journals demonstrates the wide applicability of the methods.
Despite its importance, MR-based analyses are too often poorly conducted and poorly
reported. I believe one reason for this incongruity is inconsistency between how material is
presented and how most students best learn.
Anyone who teaches (or has ever taken) courses in statistics and research methodology
knows that many students, even those who may become gifted researchers, do not always
gain conceptual understanding through numerical presentation. Although many who teach
statistics understand the processes underlying a sequence of formulas and gain conceptual
understanding through these formulas, many students do not. Instead, such students often
need a thorough conceptual explanation to gain such understanding, after which a numeri-
cal presentation may make more sense. Unfortunately, many multiple regression textbooks
assume that students will understand multiple regression best by learning matrix algebra,
wading through formulas, and focusing on details.
At the same time, methods such as structural equation modeling (SEM) and confirma-
tory factor analysis (CFA) are easily taught as extensions of multiple regression. If structured
properly, multiple regression flows naturally into these more complex topics, with nearly
complete carry-over of concepts. Path models (simple SEMs) illustrate and help deal with
some of the problems of MR, CFA does the same for path analysis, and latent variable SEM
combines all the previous topics into a powerful, flexible methodology.
I have taught courses including these topics at four universities (the University of Iowa,
Virginia Polytechnic Institute & State University, Alfred University, and the University of

vii
viii • PREFACE

Texas). These courses included faculty and students in architecture, engineering, educational
psychology, educational research and statistics, kinesiology, management, political science,
psychology, social work, and sociology, among others. This experience leads me to believe
that it is possible to teach these methods by focusing on the concepts and purposes of MR
and related methods, rather than the derivation and calculation of formulas (what my wife
calls the “plug and chug” method of learning statistics). Students generally find such an
approach clearer, more conceptual, and less threatening than other approaches. As a result
of this conceptual approach, students become interested in conducting research using MR,
CFA, or SEM and are more likely to use the methods wisely.

THE ORIENTATION OF THIS BOOK

My overriding bias in this book is that these complex methods can be presented and learned
in a conceptual, yet rigorous, manner. I recognize that not all topics are covered in the depth
or detail presented in other texts, but I will direct you to other sources for topics for which
you may want additional detail. My style is also fairly informal; I’ve written this book as if I
were teaching a class.

Data
I also believe that one learns these methods best by doing, and the more interesting and rel-
evant that “doing,” the better. For this reason, there are numerous example analyses through-
out this book that I encourage you to reproduce as you read. To make this task easier, the Web
site that accompanies the book (www.tzkeith.com) includes the data in a form that can be
used in common statistical analysis programs. Many of the examples are taken from actual
research in the social sciences, and I’ve tried to sample from research from a variety of areas.
In most cases simulated data are provided that mimic the actual data used in the research.
You can reproduce the analyses of the original researchers and, perhaps, improve on them.
And the data feast doesn’t end there! The Web site also includes data from a major federal
data set: 1000 cases from the National Education Longitudinal Study (NELS) from the National
Center for Education Statistics. NELS was a nationally representative sample of 8th-grade stu-
dents first surveyed in 1988 and resurveyed in 10th and 12th grades and then twice after leav-
ing high school. The students’ parents, teachers, and school administrators were also surveyed.
The Web site includes student and parent data from the base year (8th grade) and student data
from the first follow-up (10th grade). Don’t be led astray by the word Education in NELS; the
students were asked an incredible variety of questions, from drug use to psychological well-
being to plans for the future. Anyone with an interest in youth will find something interesting
in these data. Appendix A includes more information about the data at www.tzkeith.com.

Computer Analysis
Finally, I firmly believe that any book on statistics or research methods should be closely related
to statistical analysis software. Why plug and chug—plug numbers into formulas and chug out
the answers on a calculator—when a statistical program can do the calculations more quickly
and accurately with, for most people, no loss of understanding? Freed from the drudgery of
hand calculations, you can then concentrate on asking and answering important research ques-
tions, rather than on the intricacies of calculating statistics. This bias toward computer calcu-
lations is especially important for the methods covered in this book, which quickly become
unmanageable by hand. Use a statistical analysis program as you read this book; do the exam-
ples with me and the problems at the end of the chapters, using that program.
Which program? I use SPSS as my general statistical analysis program, and you can get
the program for a reasonable price as a student in a university (approximately $100–$125
 PREFACE • ix

per year for the “Grad Pack” as this is written). But you need not use SPSS; any of the com-
mon packages will do (e.g., SAS or SYSTAT). The output in the text has a generic look to it,
which should be easily translatable to any major statistical package output. In addition, the
website (www.tzkeith.com) includes sample multiple regression and SEM output from vari-
ous statistical packages.
For the second half of the book, you will need access to a structural equation modeling
program. Fortunately, student or tryout versions of many such programs are available online.
Student pricing for the program used extensively in this book, Amos, is available, at this writ-
ing, for approximately $50 per year as an SPSS add-on. Although programs (and pricing)
change, one current limitation of Amos is that there is no Mac OS version of Amos. If you
want to use Amos, you need to be able to run Windows. Amos is, in my opinion, the easiest
SEM program to use (and it produces really nifty pictures). The other SEM program that
I will frequently reference is Mplus. We’ll talk more about SEM in Part 2 of this book. The
website for this text has many examples of SEM input and output using Amos and Mplus.

Overview of the Book

This book is divided into two parts. Part 1 focuses on multiple regression analysis. We begin
by focusing on simple, bivariate regression and then expand that focus into multiple regres-
sion with two, three, and four independent variables. We will concentrate on the analysis
and interpretation of multiple regression as a way of answering interesting and important
research questions. Along the way, we will also deal with the analytic details of multiple
regression so that you understand what is going on when we do a multiple regression analysis.
We will focus on three different types, or flavors, of multiple regression that you will encoun-
ter in the research literature, their strengths and weaknesses, and their proper interpretation.
Our next step will be to add categorical independent variables to our multiple regression
analyses, at which point the relation of multiple regression and ANOVA will become clearer.
We will learn how to test for interactions and curves in the regression line and to apply these
methods to interesting research questions.
The penultimate chapter for Part 1 is a review chapter that summarizes and integrates
what we have learned about multiple regression. Besides serving as a review for those who
have gone through Part 1, it also serves as a useful introduction for those who are interested
primarily in the material in Part 2. In addition, this chapter introduces several important
topics not covered completely in previous chapters. The final chapter in Part 1 presents two
related methods, logistic regression and multilevel modeling, in a conceptual fashion using
what we have learned about multiple regression.
Part 2 focuses on structural equation modeling—the “Beyond” portion of the book’s
title. We begin by discussing path analysis, or structural equation modeling with measured
variables. Simple path analyses are easily estimated via multiple regression analysis, and
many of our questions about the proper use and interpretation of multiple regression will
be answered with this heuristic aid. We will deal in some depth with the problem of valid
versus invalid inferences of causality in these chapters. The problem of error (“the scourge of
research”) serves as our jumping off place for the transition from path analysis to methods
that incorporate latent variables (confirmatory factor analysis and latent variable structural
equation modeling). Confirmatory factor analysis (CFA) approaches more closely the con-
structs of primary interest in our research by separating measurement error from variation
due to these constructs. Latent variable structural equation modeling (SEM) incorporates
the advantages of path analysis with those of confirmatory factor analysis into a powerful
and flexible analytic system that partially obviates many of the problems we discuss as the
book progresses. As we progress to more advanced SEM topics we will learn how to test for
x • PREFACE

interactions in SEM models, and for differences in means of latent constructs. SEM allows
powerful analysis of change over time via methods such as latent growth models. Even when
we discuss fairly sophisticated SEMs, we reiterate one more time the possible dangers of
nonexperimental research in general and SEM in particular.

CHANGES TO THE SECOND EDITION

If you are coming to the second edition from the first, thank you! There are changes through-
out the book, including quite a few new topics, especially in Part 2. Briefly, these include:

Changes to Part 1
All chapters have been updated to add, I hope, additional clarity. In some chapters the exam-
ples used to illustrate particular points have been replaced with new ones. In most chapters I
have added additional exercises and have tried to sample these from a variety of disciplines.
New to Part 1 is a chapter on Logistic Regression and Multilevel Modeling (Chapter 10). This
brief introduction is not intended as an introduction to these important topics but instead as
a bridge to assist students who are interested in pursuing these topics in more depth in subse-
quent coursework. When I teach MR classes I consistently get questions about these methods,
how to think about them, and where to go for more information. The chapter focuses on using
what students have learned so far in MR, especially categorical variables and interactions, to
bridge the gap between a MR class and ones that focus in more detail on LR and MLM.

Changes to Part 2
What is considered introductory material in SEM has expanded a great deal since I wrote the
first edition to Multiple Regression and Beyond, and thus new chapters have been added to
address these additional topics.
A chapter on Latent Means in SEM (Chapter 18) introduces the topic of mean structures
in SEM, which is required for understanding the next three chapters and which has increas-
ingly become a part of introductory classes in SEM. The chapter uses a research example to
illustrate two methods of incorporating mean structures in SEM: MIMIC-type models and
multi-group mean and covariance structure models.
A second chapter on Confirmatory Factor Analysis has been added (Chapter 19). Now
that latent means have been introduced, this chapter revisits CFA, with the addition of latent
means. The topic of invariance testing across groups, hinted at in previous chapters, is cov-
ered in more depth.
Chapter 20 focuses on Latent Growth Models. Longitudinal models and data have been
covered in several places in the text. Here latent growth models are introduced as a method
of more directly studying the process of change.
Along with these additions, Chapter 17 (Latent Variable Models: More Advanced Topics)
and the final SEM summary chapter (Chapter 21) have been extensively modified as well.

Changes to the Appendices

Appendix A, which focused on the data sets used for the text, is considerably shortened, with
the majority of the material transferred to the web (www.tzkeith.com). Likewise, the infor-
mation previously contained in appendices illustrating output from statistics programs and
SEM programs has been transferred to the web, so that I can update it regularly. There are
still appendices focused on a review of basic statistics (Appendix B) and on understanding
partial and semipartial correlations (Appendix C). The tables showing the symbols used in
the book and useful formulae are now included in appendices as well.
 Acknowledgments

This project could not have been completed without the help of many people. I was amazed
by the number of people who wrote to me about the first edition with questions, compli-
ments, and suggestions (and corrections!). Thank you! I am very grateful to the students
who have taken my classes on these topics over the years. Your questions and comments have
helped me understand what aspects of the previous edition of the book worked well and
which needed improvement or additional explanation. I owe a huge debt to the former and
current students who “test drove” the new chapters in various forms.
I am grateful to the colleagues and students who graciously read and commented on vari-
ous new sections of the book: Jacqueline Caemmerer, Craig Enders, Larry Greil, and Keenan
Pituch. I am especially grateful to Matthew Reynolds, who read and commented on every
one of the new chapters and who is a wonderful source of new ideas for how to explain dif-
ficult concepts.
I thank my hard-working editor, Rebecca Novack, and her assistants at Routledge for all
of their assistance. Rebecca’s zest and humor, and her commitment to this project, were key
to its success. None of these individuals is responsible for any remaining deficiencies of the
book, however.
Finally, a special thank you to my wife and to my sons and their families. Davis, Scotty, and
Willie, you are a constant source of joy and a great source of research ideas! Trisia provided
advice, more loving encouragement than I deserve, and the occasional nudge, all as needed.
Thank you, my love, I really could not have done this without you!

xi
This page intentionally left blank
 Part I
Multiple Regression

1
This page intentionally left blank
 1
Introduction
Simple (Bivariate) Regression

Simple (Bivariate) Regression 4

Example: Homework and Math Achievement 4
Regression in Perspective 15
Relation of Regression to Other Statistical Methods 15
Explaining Variance 17
Advantages of Multiple Regression 18
Other Issues 19
Prediction Versus Explanation 19
Causality19
Review of Some Basics 20
Variance and Standard Deviation 20
Correlation and Covariance 20
Working With Extant Data Sets 21
Summary23
Exercises24
Notes24

This book is designed to provide a conceptually oriented introduction to multiple regres­

sion along with more complex methods that flow naturally from multiple regression: path
analysis, confirmatory factor analysis, and structural equation modeling. In this introduc­
tory chapter, we begin with a discussion and example of simple, or bivariate, regression. For
many readers, this will be a review, but, even then, the example and computer output should
provide a transition to subsequent chapters and to multiple regression. The chapter also
reviews several other related concepts, and introduces several issues (prediction and expla­
nation, causality) that we will return to repeatedly in this book. Finally, the chapter relates
regression to other approaches with which you may be more familiar, such as analysis of
variance (ANOVA). I will demonstrate that ANOVA and regression are fun�damentally the
same process and that, in fact, regression subsumes ANOVA.
As I suggested in the Preface, we start this journey by jumping right into an example and
explaining it as we go. In this introduction, I have assumed that you are fairly familiar with
the topics of correlation and statistical significance testing and that you have some familiar­
ity with statistical procedures such as the t test for comparing means and analysis of vari­
ance. If these concepts are not familiar to you a quick review is provided in Appendix B. This

3
4 • MULTIPLE REGRESSION

appendix reviews basic statistics, distributions, standard errors and confidence intervals,
correlations, t tests, and ANOVA.

SIMPLE (BIVARIATE) REGRESSION

Let’s start our adventure into the wonderful world of multiple regression with a review of sim­
ple, or bivariate, regression; that is, regression with only one influence (independent variable)
and one outcome (dependent variable).1 Pretend that you are the parent of an adolescent.
As a parent, you are interested in the influences on adolescents’ school performance: what’s
important and what’s not? Homework is of particular interest because you see your daughter
Lisa struggle with it nightly and hear her complain about it daily. A quick search of the Internet
reveals conflicting evidence. You may find books (Kohn, 2006) and articles (Wallis, 2006) criti­
cal of homework and homework policies. On the other hand, you may find links to research
suggesting homework improves learning and achievement (Cooper, Robinson, & Patall, 2006).
So you wonder if homework is just busywork or is it a worthwhile learning experience?

Example: Homework and Math Achievement

The Data
Fortunately for you, your good friend is an 8th-grade math teacher and you are a researcher;
you have the means, motive, and opportunity to find the answer to your question. Without
going into the levels of permission you’d need to collect such data, pretend that you devise a
quick survey that you give to all 8th-graders. The key question on this survey is:
Think about your math homework over the last month. Approximately how much time
did you spend, per week, doing your math homework? Approximately ____ (fill in the blank)
hours per week.
A month later, standardized achievement tests are administered; when they are available,
you record the math achievement test score for each student. You now have a report of aver­
age amount of time spent on math homework and math achievement test scores for 100
8th-graders.
A portion of the data is shown in Figure 1.1. The complete data are on the website that
accompanies this book, www.tzkeith.com, under Chapter 1, in several formats: as an SPSS
System file (homework & ach.sav), as a Microsoft Excel file (homework & ach.xls), and as an
ASCII, or plain text, file (homework & ach.txt). The values for time spent on Math Home­
work are in hours, ranging from zero for those who do no math homework to some upper
value limited by the number of free hours in a week. The Math Achievement test scores have
a national mean of 50 and a standard deviation of 10 (these are known as T scores, which
have nothing to do with t tests).2
Let’s turn to the analysis. Fortunately, you have good data analytic habits: you check basic
descriptive data prior to doing the main regression analysis. Here’s my rule: Always, always,
always, always, always, always check your data prior to conducting analyses! The frequencies
and descrip�tive statistics for the Math Homework variable are shown in Figure 1.2. Reported
Math Home�work ranged from no time, or zero hours, reported by 19 students, to 10 hours
per week. The range of values looks reasonable, with no excessively high or impossible val­
ues. For example, if someone had reported spending 40 hours per week on Math Homework,
you might be a lit�tle suspicious and would check your original data to make sure you entered the
data correctly (e.g., you may have entered a “4” as a “40”). You might be a little surprised that
the average amount of time spent on Math Homework per week is only 2.2 hours, but this
value is certainly plausible. (As noted in the Preface, the regression and other results shown
 Math Homework Math Achievement
2 54
0 53
4 53
0 56
2 59
0 30
1 49
0 54
3 37
0 49
4 55
7 50
3 45
1 44
1 60
0 36
3 53
0 22
1 56
(Data Continue............................)
Figure 1.1╇ Portion of the Math Homework and Achievement data. The complete data are on the
website under Chapter 1.

MATHHOME Time Spent on Math Homework per Week

Cumulative
Frequency Percent Valid Percent Percent
Valid .00 19 19.0 19.0 19.0
1.00 19 19.0 19.0 38.0
2.00 25 25.0 25.0 63.0
3.00 16 16.0 16.0 79.0
4.00 11 11.0 11.0 90.0
5.00 6 6.0 6.0 96.0
6.00 2 2.0 2.0 98.0
7.00 1 1.0 1.0 99.0
10.00 1 1.0 1.0 100.0
Total 100 100.0 100.0

Statistics
MATHHOME Time Spent on Math Homework per Week
N Valid 100
Missing 0
Mean 2.2000
Median 2.0000
Mode 2.00
Std. Deviation 1.8146
Variance 3.2929
Minimum .00
Maximum 10.00
Sum 220.00
Figure 1.2╇ Frequencies and descriptive statistics for Math Homework.
6 • MULTIPLE REGRESSION

are portions of an SPSS printout, but the information displayed is easily generalizable to that
produced by other statistical programs.)
Next, turn to the descriptive statistics for the Math Achievement test (Figure 1.3). Again,
given that the national mean for this test is 50, the 8th-grade school mean of 51.41 is reason­
able, as is the range of scores from 22 to 75. In contrast, if the descriptive statistics had shown
a high of, for example, 90 (four standard deviations above the mean), further investigation
would be called for. The data appear to be in good shape.

The Regression Analysis

Next, we conduct regression: we regress Math Achievement scores on time spent on Homework
(notice the structure of this statement: we regress the outcome on the influence or influences).
Figure 1.4 shows the means, standard deviations, and correlation between the two variables.

Descriptive Statistics

N Range Minimum Maximum Sum Mean Std. Deviation Variance

MATHACH Math
100 53.00 22.00 75.00 5141.00 51.4100 11.2861 127.376
Achievement Test Score
Valid N (listwise) 100

Figure 1.3╇ Descriptive statistics for Math Achievement test scores.

Descriptive Statistics

Mean Std. Deviation N

MATHACH Math
Achievement Test Score 51.4100 11.2861 100
MATHHOME Time
Spent on Math 2.2000 1.8146 100
Homework per Week

Correlations

MATHHOME
MATHACH Time Spent
Math on Math
Achievement Homework
Test Score per Week
Pearson Correlation MATHACH Math
Achievement Test Score 1.000 .320
MATHHOME Time
Spent on Math .320 1.000
Homework per Week
Sig. (1-tailed) MATHACH Math
Achievement Test Score . .001
MATHHOME Time
Spent on Math .001 .
Homework per Week
N MATHACH Math
Achievement Test Score 100 100
MATHHOME Time
Spent on Math 100 100
Homework per Week

Figure 1.4╇ Results of the regression of Math Achievement on Math Homework: descriptive statistics
and correlation coefficients.
 INTRODUCTION: SIMPLE (BIVARIATE) REGRESSION • 7

The descriptive statistics match those presented earlier, without the detail. The corre�lation
between the two variables is .320, not large, but certainly statistically significant (p€<€.01)
with this sample of 100 students. As you read articles that use multiple regression, you may
see this ordinary correlation coefficient referred to as a zero-order correlation (which dis­
tinguishes it from first-, second-, or multiple-order partial correlations, topics dis�cussed in
Appendix C).
Next, we turn to the regression itself; although we have conducted a simple regres�sion, the
computer output is in the form of multiple regression to allow a smooth transition. First,
look at the model summary in Figure 1.5. It lists the R, which normally is used to des�ignate
the multiple correlation coefficient, but which, with one predictor, is the same as the simple
Pearson correlation (.320).3 Next is the R2, which denotes the variance explained in the out­
come variable by the predictor variables. Homework time explains, accounts for, or predicts
.102 (proportion) or 10.2% of the variance in Math test scores. As you run this regression
yourself, your output will probably show some additional statistics (e.g., the adjusted R2); we
will ignore these for the time being.
Is the regression, that is, the multiple R and R2, statistically significant? We know it is,
because we already noted the statistical significance of the zero-order correlation, and this
“multiple” regression is actually a simple regression with only one predictor. But, again, we’ll
check the output for consistency with subsequent examples. Interestingly, we use an F test, as
in ANOVA, to test the statistical significance of the regression equation:
ssregression / df regression
F=
ssresidual / df residual

The term ssregression stands for sums of squares regression and is a measure of the variation
in the dependent variable that is explained by the independent variable(s); the ssresidual is the
vari�ance unexplained by the regression. If you are interested in knowing how to calculate
these values by hand, turn to Note 4 at the end of this chapter; here, we will use the values
from the statistical output in Figure 1.5.4 The sums of squares for the regression versus the

Model Summary

Model R R Square
1 .320a .102
a. Predictors: (Constant), MATHHOME Time
Spent on Math Homework per Week

ANOVAb

Sum of
Model Squares df Mean Square F Sig.
1 Regression 1291.231 1 1291.231 11.180 .001a
Residual 11318.959 98 115.500
Total 12610.190 99
a. Predictors: (Constant), MATHHOME Time Spent on Math Homework per Week
b. Dependent Variable: MATHACH Math Achievement Test Score
Figure 1.5╇ Results of the regression of Math Achievement on Math Homework: statistical significance
of the regression.
8 • MULTIPLE REGRESSION

residual are shown in the ANOVA table. In regression, the degrees of freedom (df) for the
regression are equal to the number of independent variables (k), and the df for the residual,
or error, are equal to the sample size minus the number of independent variables in the equa­
tion minus 1 (N€−€k€−€1); the df are also shown in the ANOVA table. We’ll double-check the
numbers:
1291.231 / 1
F=
11318.959 / 98
1291.231
=
115.500
= 11.179
which is the same value shown in the table, within errors of rounding. What is the probabil­
ity of obtaining a value of F as large as 11.179 if these two variables were in fact unrelated
in the population? According to the table (in the column labeled “Sig.”), such an occurrence
would occur only 1 time in 1,000 (p = .001); it would seem logical that these two variables
are indeed related. We can double-check this probability by referring to an F table under 1
and 98 df; is the value 11.179 greater than the tabled value? Instead, however, I suggest that
you use a computer program to calculate these probabilities. Excel, for example, will find the
probability for values of all the distributions discussed in this text. Simply put the calculated
value of F (11.179) in one cell, the degrees of freedom for the regression (1)€in the next,
and€the df for the residual in the next (98). Go to the next cell, then click on Insert, Function,
and select the category of Statistical and scroll down until you find FDIST, for F distribution.
Click on it and point to the cells containing the required information. Alternatively, you
could go directly to Function and FDIST and simply type in these numbers, as was done in
Figure 1.6. Excel returns a value of .001172809, or .001, as shown in the Figure. Although I
present this method of determining probabilities as a way of double-checking the computer
output at this point, at times your computer program will not display the probabilities you
are interested in, and this method will be useful.

Figure 1.6╇ Using Excel to calculate probability: statistical significance of an F (1,98) of 11.179.
 INTRODUCTION: SIMPLE (BIVARIATE) REGRESSION • 9

There is another formula you can use to calculate F, an extension of which will come in
handy later:

R2 / k
F=
(1− R 2 ) /(N − k −1)

This formula compares the proportion of variance explained by the regression (R2) with
the proportion of variance left unexplained by the regression (1€−€R2). This formula may
seem quite different from the one presented previously until you remember that (1) k is
equal to the df for the regression, and N€−€k€−€1 is equal to the df for the residual, and (2) the
sums of squares from the previous formula are also estimates of variance. Try this formula to
make sure you get the same results (within rounding error).
I noted that the ssregression is a measure of the variance explained in the dependent variable
by the independent variables, and also that R2 denotes the variance explained. Given these
descriptions, you may expect that the two ss
concepts should be related. They are, and we can
calculate the R2 from the ssregression: R 2 = regression
sstotal . We can put this formula into words: There is
a certain amount of variance in the dependent variable (total variance), and the independent
variables can explain a portion of this variance (variance due to the regression). The R2 is a
proportion of the total variance in the dependent variable that is explained by the indepen­
dent variables. For the current example, the total variance in the dependent variable, Math
Achievement (sstotal), was 12610.190 (Figure 1.5), and Math Homework explained 1291.231
of this variance. Thus,
ssregression
R2 =
sstotal
1291.231
=
12610.190
= .102
and Homework explains .102 or 10.2% of the variance in Math Achievement. Obviously, R2
can vary between 0 (no variance explained) and 1 (100% explained).

The Regression Equation

Next, let’s take a look at the coefficients for the regression equation, the notable parts
of which are shown in Figure 1.7. The general formula for a regression equation is Y =
a€+€bX€+€e, which, translated into English, says that a person’s score on the dependent vari­
able (in this case, Math Achievement) is a result of a con�stant (a), plus a coefficient (b)
times his or her value on the independent variable (Math Homework), plus error. Values
for both a and b are shown in the second column of the table in Figure 1.7 (Unstan­
dardized Coefficients, B; SPSS uses the uppercase B rather than the lower case b). a is
a constant, called the intercept, and its value is 47.032 for this homework–achievement
example. The intercept is the predicted score on the dependent variable for someone with a
score of zero on the independent variable. b, the unstandard�ized regression coefficient, is
1.990. Because we don’t have a direct estimate of the error, we’ll focus on a different form
of the regression equation: Y' = a€+€bX, in which Y' is the preÂ�dicted value of Y. The com­
pleted equation is Y' = 47.032€+€1.990X, meaning that to predict a person’s Math Achieve­
ment score we can multiply his or her report of time spent on Math Homework by 1.990
and add 47.032. Thus, the predicted score for a student who does no homework would
be 47.032, the predicted score for an 8th-grader who does 1 hour of homework is 49.022
10 • MULTIPLE REGRESSION

Coefficientsa

Unstandardized Standardized
Coefficients Coefficients 95% Confidence Interval for B
Model B Std. Error Beta t Sig. Lower Bound Upper Bound
1 Intercept (Constant) 47.032 1.694 27.763 .000 43.670 50.393
MATHHOME Time
Spent on Math 1.990 .595 .320 3.344 .001 .809 3.171
Homework per Week
a. Dependent Variable: MATHACH Math Achievement Test Score
Figure 1.7╇Results of the regression of Math Achievement on Math Homework: Regression
Coefficients.

(1€×€1.990€+€47.032), the predicted score for a student who does 2 hours of homework is
51.012 (2 ×€1.990€+€47.032), and so on.
Several questions may spring to mind after these last statements. Why, for example, would
we want to predict a student’s Achievement score (Y') when we already know the student’s
real Achievement score? The answer is that we want to use this formula to summarize the
relation between homework and achievement for all students at the same time. We may also
be able to use the formula for other purposes: to predict scores for another group of students
or, to return to the original purpose, to predict Lisa’s likely future math achievement, given
her time spent on math homework. Or we may want to know what would likely happen if
a student or group of students were to increase or decrease the time they spent on math
homework.

Interpretation
But to get back to our original question, we now have some very useful information for Lisa,
contained within the regression coefficient (b€=€1.99), because this coefficient tells us the
amount we can expect the outcome variable (Math Achievement) to change for each 1-unit
change in the independent variable (Math Homework). Because the Homework variable is
in hours spent per week, we can make this statement: “For each addiÂ�tional hour students
spend on Mathematics Homework every week, they can expect to see close to a 2-point
increase in Math Achievement test scores.” Now, Achievement test scores are not that easy to
change; it is much easier, for example, to improve grades than test scores (Keith, Diamond-
Hallam, & Fine, 2004), so this represents an important effect. Given the standard deviation
of the test scores (10 points), a student should be able to improve his or her scores by a stan­
dard deviation by studying a little more than 5 extra hours a week; this could mean moving
from average-level to high-average-level achievement. Of course, this proposition might be
more interesting to a student who is currently spending very little time studying than to one
who is already spending a lot of time working on math homework.

The Regression Line

The regression equation may be used to graph the relation between Math Homework and
Achievement, and this graph can also illustrate nicely the predictions made in the previous
paragraph. The intercept (a) is the value on the Y (Achievement) axis for a value of zero for
X (Homework); in other words, the intercept is the value on the Achievement test we would
expect for someone who does no homework. We can use the intercept as one data point for
drawing the regression line (X = 0, Y = 47.032). The second data point is simply the point
defined by the mean of X (Mx = 2.200) and the mean of Y (My = 51.410). The graph, with
these two data points highlighted, is shown in Figure 1.8. We can use the graph and data to
 INTRODUCTION: SIMPLE (BIVARIATE) REGRESSION • 11

Regression of Math Achievement on Math Homework

80

70
Achievement Test Scores

60

50
51.41
47.032
40

30

20
0 1 2 3 4 5 6 7 8 9 10 11 12
Time Spent on Math Homework per Week
Figure 1.8╇ Regression line for Math Achievement on Math Homework. The line is drawn through the
intercept and the joint means of X and Y.

check the calculation of the value of b, which is equal to the slope of the regression line. The
slope is equal to the increase in Y for each unit increase in X (or the rise of the line divided by
the run of the line); we can use the two data points plotted to calculate the slope:
rise M y − a
b= =
run M x − 0
51.410 − 47.032
=
2.200
= 1.990
Let’s consider for a few moments the graph and these formulas. The slope represents the
predicted increase in Y for each unit increase in X. For this example, this means that for each
unit—in this case, each hour—increase in Homework, Achievement scores increase, on aver­
age, 1.990 points. This, then, is the interpretation of an unstandardized coefficient: It is the
predicted increase in Y expected for each unit increase in X. When the independent variable
has a meaningful metric, like hours spent studying Mathematics every week, the interpre­
tation of b is easy and straightforward. We can also generalize from this group-gen�erated
equation to individuals (to the extent that they are similar to the group that generated the
regression equation). Thus the graph and b can be used to make predictions for others, such
as Lisa. She can check her current level of homework time and see how much payoff she
might expect for additional time (or how much she can expect to lose if she studies less). The
intercept is also worth noting; it shows that the average Achievement test score for stu�dents
who do no studying is 47.032, slightly below the national average.
Because we are using a modern statistical package, there is no need to draw the plot of the
regression line ourselves; any such program will do it for us. Figure 1.9 shows the data points
and regression line drawn using SPSS (a scatterplot was created using the graph feature; see
www.tzkeith.com for examples). The small circles in this figure are the actual data points;
12 • MULTIPLE REGRESSION

80.00

70.00
Math Achievement Test Score

60.00

50.00

40.00

30.00

20.00

0.00 2.00 4.00 6.00 8.00 10.00

Time Spent on Math Homework per Week
Figure 1.9╇ Regression line, with data points, as produced by the SPSS Scatter/Dot graph command.

notice how variable they are. If the R were larger, the data points would cluster more closely
around the regression line. We will return to this topic in a subsequent chapter.

Statistical Significance of Regression Coefficients

There are a few more details to study for this regression analysis before stepping back and
further considering the meaning of the results. With multiple regression, we will also be
interested in whether each regres�sion coefficient is statistically significant. Return to the table
of regression coefficients (Figure 1.7), and note the columns labeled t and Sig. The values
corresponding to the regression coefficient are simply the results of a t test of the statistical
significance of the regression coefficient (b). The formula for t is one of the most ubiquitous
in statistics (Kerlinger, 1986):

statistic
t= , or, in this case,
standard error of thestatistic
b 1.990
t= = = 3.345.
SEb .595
As shown in Figure 1.7, the value of t is 3.344, with N€−€k€−€1 degrees of freedom (98).
If we look up this value in Excel (using the function TDIST), we find the probability of
obtaining such a t by chance is .001171 (a two-tailed test) rounded off to .001 (the value
Exploring the Variety of Random
Documents with Different Content
The Project Gutenberg eBook of The People of
the Black Circle
This ebook is for the use of anyone anywhere in the United States
and most other parts of the world at no cost and with almost no
restrictions whatsoever. You may copy it, give it away or re-use it
under the terms of the Project Gutenberg License included with this
ebook or online at www.gutenberg.org. If you are not located in the
United States, you will have to check the laws of the country where
you are located before using this eBook.

Title: The People of the Black Circle

Author: Robert E. Howard

Release date: March 4, 2013 [eBook #42259]

Most recently updated: October 23, 2024

Language: English

Credits: Produced by Greg Weeks, Mary Meehan and the Online

Distributed Proofreading Team at http://www.pgdp.net

*** START OF THE PROJECT GUTENBERG EBOOK THE PEOPLE OF

THE BLACK CIRCLE ***
THE PEOPLE OF THE BLACK
CIRCLE
 By Robert E. Howard
[Transcriber's Note: This etext was first published in Weird Tales
September, October, November 1934. Extensive research did not
uncover any evidence that the U.S. copyright on this publication
was renewed.]
 1 Death Strikes a King
The king of Vendhya was dying. Through the hot, stifling night the
temple gongs boomed and the conchs roared. Their clamor was a
faint echo in the gold-domed chamber where Bunda Chand
struggled on the velvet-cushioned dais. Beads of sweat glistened on
his dark skin; his fingers twisted the gold-worked fabric beneath
him. He was young; no spear had touched him, no poison lurked in
his wine. But his veins stood out like blue cords on his temples, and
his eyes dilated with the nearness of death. Trembling slave-girls
knelt at the foot of the dais, and leaning down to him, watching him
with passionate intensity, was his sister, the Devi Yasmina. With her
was the wazam, a noble grown old in the royal court.
She threw up her head in a gusty gesture of wrath and despair as
the thunder of the distant drums reached her ears.
'The priests and their clamor!' she exclaimed. 'They are no wiser
than the leeches who are helpless! Nay, he dies and none can say
why. He is dying now—and I stand here helpless, who would burn
the whole city and spill the blood of thousands to save him.'
'Not a man of Ayodhya but would die in his place, if it might be,
Devi,' answered the wazam. 'This poison—'
'I tell you it is not poison!' she cried. 'Since his birth he has been
guarded so closely that the cleverest poisoners of the East could not
reach him. Five skulls bleaching on the Tower of the Kites can testify
to attempts which were made—and which failed. As you well know,
there are ten men and ten women whose sole duty is to taste his
food and wine, and fifty armed warriors guard his chamber as they
guard it now. No, it is not poison; it is sorcery—black, ghastly magic
—'
She ceased as the king spoke; his livid lips did not move, and there
was no recognition in his glassy eyes. But his voice rose in an eery
call, indistinct and far away, as if called to her from beyond vast,
wind-blown gulfs.
'Yasmina! Yasmina! My sister, where are you? I can not find you. All
is darkness, and the roaring of great winds!'
'Brother!' cried Yasmina, catching his limp hand in a convulsive
grasp. 'I am here! Do you not know me—'
Her voice died at the utter vacancy of his face. A low confused moan
waned from his mouth. The slave-girls at the foot of the dais
whimpered with fear, and Yasmina beat her breast in anguish.

In another part of the city a man stood in a latticed balcony

overlooking a long street in which torches tossed luridly, smokily
revealing upturned dark faces and the whites of gleaming eyes. A
long-drawn wailing rose from the multitude.
The man shrugged his broad shoulders and turned back into the
arabesque chamber. He was a tall man, compactly built, and richly
clad.
'The king is not yet dead, but the dirge is sounded,' he said to
another man who sat cross-legged on a mat in a corner. This man
was clad in a brown camel-hair robe and sandals, and a green
turban was on his head. His expression was tranquil, his gaze
impersonal.
'The people know he will never see another dawn,' this man
answered.
The first speaker favored him with a long, searching stare.
'What I can not understand,' he said, 'is why I have had to wait so
long for your masters to strike. If they have slain the king now, why
could they not have slain him months ago?'
'Even the arts you call sorcery are governed by cosmic laws,'
answered the man in the green turban. 'The stars direct these
actions, as in other affairs. Not even my masters can alter the stars.
Not until the heavens were in the proper order could they perform
this necromancy.' With a long, stained fingernail he mapped the
constellations on the marble-tiled floor. 'The slant of the moon
presaged evil for the king of Vendhya; the stars are in turmoil, the
Serpent in the House of the Elephant. During such juxtaposition, the
invisible guardians are removed from the spirit of Bhunda Chand. A
path is opened in the unseen realms, and once a point of contact
was established, mighty powers were put in play along that path.'
'Point of contact?' inquired the other. 'Do you mean that lock of
Bhunda Chand's hair?'
'Yes. All discarded portions of the human body still remain part of it,
attached to it by intangible connections. The priests of Asura have a
dim inkling of this truth, and so all nail trimmings, hair and other
waste products of the persons of the royal family are carefully
reduced to ashes and the ashes hidden. But at the urgent entreaty
of the princess of Khosala, who loved Bhunda Chand vainly, he gave
her a lock of his long black hair as a token of remembrance. When
my masters decided upon his doom, the lock, in its golden, jewel-
encrusted case, was stolen from under her pillow while she slept,
and another substituted, so like the first that she never knew the
difference. Then the genuine lock travelled by camel-caravan up the
long, long road to Peshkhauri, thence up the Zhaibar Pass, until it
reached the hands of those for whom it was intended.'
'Only a lock of hair,' murmured the nobleman.
'By which a soul is drawn from its body and across gulfs of echoing
space,' returned the man on the mat.
The nobleman studied him curiously.
'I do not know if you are a man or a demon, Khemsa,' he said at
last. 'Few of us are what we seem. I, whom the Kshatriyas know as
Kerim Shah, a prince from Iranistan, am no greater a masquerader
than most men. They are all traitors in one way or another, and half
of them know not whom they serve. There at least I have no
doubts; for I serve King Yezdigerd of Turan.'
'And I the Black Seers of Yimsha,' said Khemsa; 'and my masters are
greater than yours, for they have accomplished by their arts what
Yezdigerd could not with a hundred thousand swords.'

Outside, the moan of the tortured thousands shuddered up to the

stars which crusted the sweating Vendhyan night, and the conchs
bellowed like oxen in pain.
In the gardens of the palace the torches glinted on polished helmets
and curved swords and gold-chased corselets. All the noble-born
fighting-men of Ayodhya were gathered in the great palace or about
it, and at each broad-arched gate and door fifty archers stood on
guard, with bows in their hands. But Death stalked through the royal
palace and none could stay his ghostly tread.
On the dais under the golden dome the king cried out again, racked
by awful paroxysms. Again his voice came faintly and far away, and
again the Devi bent to him, trembling with a fear that was darker
than the terror of death.
'Yasmina!' Again that far, weirdly dreeing cry, from realms
immeasurable. 'Aid me! I am far from my mortal house! Wizards
have drawn my soul through the wind-blown darkness. They seek to
snap the silver cord that binds me to my dying body. They cluster
around me; their hands are taloned, their eyes are red like flame
burning in darkness. Aie, save me, my sister! Their fingers sear me
like fire! They would slay my body and damn my soul! What is this
they bring before me?—Aie!'

At the terror in his hopeless cry Yasmina screamed uncontrollably

and threw herself bodily upon him in the abandon of her anguish.
He was torn by a terrible convulsion; foam flew from his contorted
lips and his writhing fingers left their marks on the girl's shoulders.
But the glassy blankness passed from his eyes like smoke blown
from a fire, and he looked up at his sister with recognition.
'Brother!' she sobbed. 'Brother—'
'Swift!' he gasped, and his weakening voice was rational. 'I know
now what brings me to the pyre. I have been on a far journey and I
understand. I have been ensorcelled by the wizards of the
Himelians. They drew my soul out of my body and far away, into a
stone room. There they strove to break the silver cord of life, and
thrust my soul into the body of a foul night-weird their sorcery
summoned up from hell. Ah! I feel their pull upon me now! Your cry
and the grip of your fingers brought me back, but I am going fast.
My soul clings to my body, but its hold weakens. Quick—kill me,
before they can trap my soul for ever!'
'I cannot!' she wailed, smiting her naked breasts.
'Swiftly, I command you!' There was the old imperious note in his
failing whisper. 'You have never disobeyed me—obey my last
command! Send my soul clean to Asura! Haste, lest you damn me to
spend eternity as a filthy gaunt of darkness. Strike, I command you!
Strike!'
Sobbing wildly, Yasmina plucked a jeweled dagger from her girdle
and plunged it to the hilt in his breast. He stiffened and then went
limp, a grim smile curving his dead lips. Yasmina hurled herself face-
down on the rush-covered floor, beating the reeds with her clenched
hands. Outside, the gongs and conchs brayed and thundered and
the priests gashed themselves with copper knives.
 2 A Barbarian from the Hills
Chunder Shan, governor of Peshkhauri, laid down his golden pen
and carefully scanned that which he had written on parchment that
bore his official seal. He had ruled Peshkhauri so long only because
he weighed his every word, spoken or written. Danger breeds
caution, and only a wary man lives long in that wild country where
the hot Vendhyan plains meet the crags of the Himelians. An hour's
ride westward or northward and one crossed the border and was
among the Hills where men lived by the law of the knife.
The governor was alone in his chamber, seated at his ornately
carven table of inlaid ebony. Through the wide window, open for the
coolness, he could see a square of the blue Himelian night, dotted
with great white stars. An adjacent parapet was a shadowy line, and
further crenelles and embrasures were barely hinted at in the dim
starlight. The governor's fortress was strong, and situated outside
the walls of the city it guarded. The breeze that stirred the tapestries
on the wall brought faint noises from the streets of Peshkhauri—
occasional snatches of wailing song, or the thrum of a cithern.
The governor read what he had written, slowly, with his open hand
shading his eyes from the bronze butterlamp, his lips moving.
Absently, as he read, he heard the drum of horses' hoofs outside the
barbican, the sharp staccato of the guards' challenge. He did not
heed, intent upon his letter. It was addressed to the wazam of
Vendhya, at the royal court of Ayodhya, and it stated, after the
customary salutations:

'Let it be known to your excellency that I have faithfully carried

out your excellency's instructions. The seven tribesmen are well
guarded in their prison, and I have repeatedly sent word into
the hills that their chief come in person to bargain for their
 release. But he has made no move, except to send word that
unless they are freed he will burn Peshkhauri and cover his
saddle with my hide, begging your excellency's indulgence. This
he is quite capable of attempting, and I have tripled the
numbers of the lance guards. The man is not a native of
Ghulistan. I cannot with certainty predict his next move. But
since it is the wish of the Devi—'

He was out of his ivory chair and on his feet facing the arched door,
all in one instant. He snatched at the curved sword lying in its ornate
scabbard on the table, and then checked the movement.
It was a woman who had entered unannounced, a woman whose
gossamer robes did not conceal the rich garments beneath them any
more than they concealed the suppleness and beauty of her tall,
slender figure. A filmy veil fell below her breasts, supported by a
flowing headdress bound about with a triple gold braid and adorned
with a golden crescent. Her dark eyes regarded the astonished
governor over the veil, and then with an imperious gesture of her
white hand, she uncovered her face.
'Devi!' The governor dropped to his knees before her, surprize and
confusion somewhat spoiling the stateliness of his obeisance. With a
gesture she motioned him to rise, and he hastened to lead her to
the ivory chair, all the while bowing level with his girdle. But his first
words were of reproof.
'Your Majesty! This was most unwise! The border is unsettled. Raids
from the hills are incessant. You came with a large attendance?'
'An ample retinue followed me to Peshkhauri,' she answered. 'I
lodged my people there and came on to the fort with my maid,
Gitara.'
Chunder Shan groaned in horror.
'Devi! You do not understand the peril. An hour's ride from this spot
the hills swarm with barbarians who make a profession of murder
and rapine. Women have been stolen and men stabbed between the
fort and the city. Peshkhauri is not like your southern provinces—'
'But I am here, and unharmed,' she interrupted with a trace of
impatience. 'I showed my signet ring to the guard at the gate, and
to the one outside your door, and they admitted me unannounced,
not knowing me, but supposing me to be a secret courier from
Ayodhya. Let us not now waste time.
'You have received no word from the chief of the barbarians?'
'None save threats and curses, Devi. He is wary and suspicious. He
deems it a trap, and perhaps he is not to be blamed. The Kshatriyas
have not always kept their promises to the hill people.'
'He must be brought to terms!' broke in Yasmina, the knuckles of her
clenched hands showing white.
'I do not understand.' The governor shook his head. 'When I
chanced to capture these seven hill-men, I reported their capture to
the wazam, as is the custom, and then, before I could hang them,
there came an order to hold them and communicate with their chief.
This I did, but the man holds aloof, as I have said. These men are of
the tribe of Afghulis, but he is a foreigner from the west, and he is
called Conan. I have threatened to hang them tomorrow at dawn, if
he does not come.'
'Good!' exclaimed the Devi. 'You have done well. And I will tell you
why I have given these orders. My brother—' she faltered, choking,
and the governor bowed his head, with the customary gesture of
respect for a departed sovereign.
'The king of Vendhya was destroyed by magic,' she said at last. 'I
have devoted my life to the destruction of his murderers. As he died
he gave me a clue, and I have followed it. I have read the Book of
Skelos, and talked with nameless hermits in the caves below Jhelai. I
learned how, and by whom, he was destroyed. His enemies were the
Black Seers of Mount Yimsha.'
'Asura!' whispered Chunder Shan, paling.
Her eyes knifed him through. 'Do you fear them?'
'Who does not, Your Majesty?' he replied. 'They are black devils,
haunting the uninhabited hills beyond the Zhaibar. But the sages say
that they seldom interfere in the lives of mortal men.'
'Why they slew my brother I do not know,' she answered. 'But I
have sworn on the altar of Asura to destroy them! And I need the
aid of a man beyond the border. A Kshatriya army, unaided, would
never reach Yimsha.'
'Aye,' muttered Chunder Shan. 'You speak the truth there. It would
be fight every step of the way, with hairy hill-men hurling down
boulders from every height, and rushing us with their long knives in
every valley. The Turanians fought their way through the Himelians
once, but how many returned to Khurusun? Few of those who
escaped the swords of the Kshatriyas, after the king, your brother,
defeated their host on the Jhumda River, ever saw Secunderam
again.'
'And so I must control men across the border,' she said, 'men who
know the way to Mount Yimsha—'
'But the tribes fear the Black Seers and shun the unholy mountain,'
broke in the governor.
'Does the chief, Conan, fear them?' she asked.
'Well, as to that,' muttered the governor, 'I doubt if there is anything
that devil fears.'
'So I have been told. Therefore he is the man I must deal with. He
wishes the release of his seven men. Very well; their ransom shall be
the heads of the Black Seers!' Her voice thrummed with hate as she
uttered the last words, and her hands clenched at her sides. She
looked an image of incarnate passion as she stood there with her
head thrown high and her bosom heaving.
Again the governor knelt, for part of his wisdom was the knowledge
that a woman in such an emotional tempest is as perilous as a blind
cobra to any about her.
'It shall be as you wish, Your Majesty.' Then as she presented a
calmer aspect, he rose and ventured to drop a word of warning. 'I
can not predict what the chief Conan's action will be. The tribesmen
are always turbulent, and I have reason to believe that emissaries
from the Turanians are stirring them up to raid our borders. As your
majesty knows, the Turanians have established themselves in
Secunderam and other northern cities, though the hill tribes remain
unconquered. King Yezdigerd has long looked southward with greedy
lust and perhaps is seeking to gain by treachery what he could not
win by force of arms. I have thought that Conan might well be one
of his spies.'
'We shall see,' she answered. 'If he loves his followers, he will be at
the gates at dawn, to parley. I shall spend the night in the fortress. I
came in disguise to Peshkhauri, and lodged my retinue at an inn
instead of the palace. Besides my people, only yourself knows of my
presence here.'
'I shall escort you to your quarters, Your Majesty,' said the governor,
and as they emerged from the doorway, he beckoned the warrior on
guard there, and the man fell in behind them, spear held at salute.
The maid waited, veiled like her mistress, outside the door, and the
group traversed a wide, winding corridor, lighted by smoky torches,
and reached the quarters reserved for visiting notables—generals
and viceroys, mostly; none of the royal family had ever honored the
fortress before. Chunder Shan had a perturbed feeling that the suite
was not suitable to such an exalted personage as the Devi, and
though she sought to make him feel at ease in her presence, he was
glad when she dismissed him and he bowed himself out. All the
menials of the fort had been summoned to serve his royal guest—
though he did not divulge her identity—and he stationed a squad of
spearmen before her doors, among them the warrior who had
guarded his own chamber. In his preoccupation he forgot to replace
the man.
The governor had not been long gone from her when Yasmina
suddenly remembered something else which she had wished to
discuss with him, but had forgotten until that moment. It concerned
the past actions of one Kerim Shah, a nobleman from Iranistan, who
had dwelt for a while in Peshkhauri before coming on to the court at
Ayodhya. A vague suspicion concerning the man had been stirred by
a glimpse of him in Peshkhauri that night. She wondered if he had
followed her from Ayodhya. Being a truly remarkable Devi, she did
not summon the governor to her again, but hurried out into the
corridor alone, and hastened toward his chamber.

Chunder Shan, entering his chamber, closed the door and went to
his table. There he took the letter he had been writing and tore it to
bits. Scarcely had he finished when he heard something drop softly
onto the parapet adjacent to the window. He looked up to see a
figure loom briefly against the stars, and then a man dropped lightly
into the room. The light glinted on a long sheen of steel in his hand.
'Shhhh!' he warned. 'Don't make a noise, or I'll send the devil a
henchman!'
The governor checked his motion toward the sword on the table. He
was within reach of the yard-long Zhaibar knife that glittered in the
intruder's fist, and he knew the desperate quickness of a hillman.
The invader was a tall man, at once strong and supple. He was
dressed like a hillman, but his dark features and blazing blue eyes
did not match his garb. Chunder Shan had never seen a man like
him; he was not an Easterner, but some barbarian from the West.
But his aspect was as untamed and formidable as any of the hairy
tribesmen who haunt the hills of Ghulistan.
'You come like a thief in the night,' commented the governor,
recovering some of his composure, although he remembered that
there was no guard within call. Still, the hillman could not know that.
'I climbed a bastion,' snarled the intruder. 'A guard thrust his head
over the battlement in time for me to rap it with my knife-hilt.'
'You are Conan?'
'Who else? You sent word into the hills that you wished for me to
come and parley with you. Well, by Crom, I've come! Keep away
from that table or I'll gut you.'
'I merely wish to seat myself,' answered the governor, carefully
sinking into the ivory chair, which he wheeled away from the table.
Conan moved restlessly before him, glancing suspiciously at the
door, thumbing the razor edge of his three-foot knife. He did not
walk like an Afghuli, and was bluntly direct where the East is subtle.
'You have seven of my men,' he said abruptly. 'You refused the
ransom I offered. What the devil do you want?'
'Let us discuss terms,' answered Chunder Shan cautiously.
'Terms?' There was a timbre of dangerous anger in his voice. 'What
do you mean? Haven't I offered you gold?'
Chunder Shan laughed.
'Gold? There is more gold in Peshkhauri than you ever saw.'
'You're a liar,' retorted Conan. 'I've seen the suk of the goldsmiths in
Khurusun.'
'Well, more than an Afghuli ever saw,' amended Chunder Shan. 'And
it is but a drop of all the treasure of Vendhya. Why should we desire
gold? It would be more to our advantage to hang these seven
thieves.'
Conan ripped out a sulfurous oath and the long blade quivered in his
grip as the muscles rose in ridges on his brown arm.
'I'll split your head like a ripe melon!'
A wild blue flame flickered in the hillman's eyes, but Chunder Shan
shrugged his shoulders, though keeping an eye on the keen steel.
'You can kill me easily, and probably escape over the wall afterward.
But that would not save the seven tribesmen. My men would surely
hang them. And these men are headmen among the Afghulis.'
'I know it,' snarled Conan. 'The tribe is baying like wolves at my
heels because I have not procured their release. Tell me in plain
words what you want, because, by Crom! if there's no other way, I'll
raise a horde and lead it to the very gates of Peshkhauri!'
Looking at the man as he stood squarely, knife in fist and eyes
glaring, Chunder Shan did not doubt that he was capable of it. The
governor did not believe any hill-horde could take Peshkhauri, but he
did not wish a devastated countryside.
'There is a mission you must perform,' he said, choosing his words
with as much care as if they had been razors. 'There—'
Conan had sprung back, wheeling to face the door at the same
instant, lips asnarl. His barbarian ears had caught the quick tread of
soft slippers outside the door. The next instant the door was thrown
open and a slim, silk-robed form entered hastily, pulling the door
shut—then stopping short at sight of the hillman.
Chunder Shan sprang up, his heart jumping into his mouth.
'Devi!' he cried involuntarily, losing his head momentarily in his
fright.
'Devi!' It was like an explosive echo from the hillman's lips. Chunder
Shan saw recognition and intent flame up in the fierce blue eyes.
The governor shouted desperately and caught at his sword, but the
hillman moved with the devastating speed of a hurricane. He sprang,
knocked the governor sprawling with a savage blow of his knife-hilt,
swept up the astounded Devi in one brawny arm and leaped for the
window. Chunder Shan, struggling frantically to his feet, saw the
man poise an instant on the sill in a flutter of silken skirts and white
limbs that was his royal captive, and heard his fierce, exultant snarl:
'Now dare to hang my men!' and then Conan leaped to the parapet
and was gone. A wild scream floated back to the governor's ears.
'Guard! Guard!' screamed the governor, struggling up and running
drunkenly to the door. He tore it open and reeled into the hall. His
shouts re-echoed along the corridors, and warriors came running,
gaping to see the governor holding his broken head, from which the
blood streamed.
'Turn out the lancers!' he roared. 'There has been an abduction!'
Even in his frenzy he had enough sense left to withhold the full
truth. He stopped short as he heard a sudden drum of hoofs outside,
a frantic scream and a wild yell of barbaric exultation.
Followed by the bewildered guardsmen, the governor raced for the
stair. In the courtyard of the fort a force of lancers stood by saddled
steeds, ready to ride at an instant's notice. Chunder Shan led his
squadron flying after the fugitive, though his head swam so he had
to hold with both hands to the saddle. He did not divulge the identity
of the victim, but said merely that the noblewoman who had borne
the royal signet-ring had been carried away by the chief of the
Afghulis. The abductor was out of sight and hearing, but they knew
the path he would strike—the road that runs straight to the mouth of
the Zhaibar. There was no moon; peasant huts rose dimly in the
starlight. Behind them fell away the grim bastion of the fort, and the
towers of Peshkhauri. Ahead of them loomed the black walls of the
Himelians.
 3 Khemsa Uses Magic
In the confusion that reigned in the fortress while the guard was
being turned out, no one noticed that the girl who had accompanied
the Devi slipped out the great arched gate and vanished in the
darkness. She ran straight for the city, her garments tucked high.
She did not follow the open road, but cut straight through fields and
over slopes, avoiding fences and leaping irrigation ditches as surely
as if it were broad daylight, and as easily as if she were a trained
masculine runner. The hoof-drum of the guardsmen had faded away
up the hill before she reached the city wall. She did not go to the
great gate, beneath whose arch men leaned on spears and craned
their necks into the darkness, discussing the unwonted activity about
the fortress. She skirted the wall until she reached a certain point
where the spire of the tower was visible above the battlements.
Then she placed her hands to her mouth and voiced a low weird call
that carried strangely.
Almost instantly a head appeared at an embrasure and a rope came
wriggling down the wall. She seized it, placed a foot in the loop at
the end, and waved her arm. Then quickly and smoothly she was
drawn up the sheer stone curtain. An instant later she scrambled
over the merlons and stood up on a flat roof which covered a house
that was built against the wall. There was an open trap there, and a
man in a camel-hair robe who silently coiled the rope, not showing
in any way the strain of hauling a full-grown woman up a forty-foot
wall.
'Where is Kerim Shah?' she gasped, panting after her long run.
'Asleep in the house below. You have news?'
'Conan has stolen the Devi out of the fortress and carried her away
into the hills!' She blurted out her news in a rush, the words
stumbling over one another.
Khemsa showed no emotion, but merely nodded his turbaned head.
'Kerim Shah will be glad to hear that,' he said.
'Wait!' The girl threw her supple arms about his neck. She was
panting hard, but not only from exertion. Her eyes blazed like black
jewels in the starlight. Her upturned face was close to Khemsa's, but
though he submitted to her embrace, he did not return it.
'Do not tell the Hyrkanian!' she panted. 'Let us use this knowledge
ourselves! The governor has gone into the hills with his riders, but
he might as well chase a ghost. He has not told anyone that it was
the Devi who was kidnapped. None in Peshkhauri or the fort knows
it except us.'
'But what good does it do us?' the man expostulated. 'My masters
sent me with Kerim Shah to aid him in every way—'
'Aid yourself!' she cried fiercely. 'Shake off your yoke!'
'You mean—disobey my masters?' he gasped, and she felt his whole
body turn cold under her arms.
'Aye!' she shook him in the fury of her emotion. 'You too are a
magician! Why will you be a slave, using your powers only to elevate
others? Use your arts for yourself!'
'That is forbidden!' He was shaking as if with an ague. 'I am not one
of the Black Circle. Only by the command of the masters do I dare to
use the knowledge they have taught me.'
'But you can use it!' she argued passionately. 'Do as I beg you! Of
course Conan has taken the Devi to hold as hostage against the
seven tribesmen in the governor's prison. Destroy them, so Chunder
Shan can not use them to buy back the Devi. Then let us go into the
mountains and take her from the Afghulis. They can not stand
against your sorcery with their knives. The treasure of the Vendhyan
kings will be ours as ransom—and then when we have it in our
hands, we can trick them, and sell her to the king of Turan. We shall
have wealth beyond our maddest dreams. With it we can buy
warriors. We will take Khorbhul, oust the Turanians from the hills,
and send our hosts southward; become king and queen of an
empire!'
Khemsa too was panting, shaking like a leaf in her grasp; his face
showed gray in the starlight, beaded with great drops of
perspiration.
'I love you!' she cried fiercely, writhing her body against his, almost
strangling him in her wild embrace, shaking him in her abandon. 'I
will make a king of you! For love of you I betrayed my mistress; for
love of me betray your masters! Why fear the Black Seers? By your
love for me you have broken one of their laws already! Break the
rest! You are as strong as they!'
A man of ice could not have withstood the searing heat of her
passion and fury. With an inarticulate cry he crushed her to him,
bending her backward and showering gasping kisses on her eyes,
face and lips.
'I'll do it!' His voice was thick with laboring emotions. He staggered
like a drunken man. 'The arts they have taught me shall work for
me, not for my masters. We shall be rulers of the world—of the
world—'
'Come then!' Twisting lithely out of his embrace, she seized his hand
and led him toward the trap-door. 'First we must make sure that the
governor does not exchange those seven Afghulis for the Devi.'
He moved like a man in a daze, until they had descended a ladder
and she paused in the chamber below. Kerim Shah lay on a couch
motionless, an arm across his face as though to shield his sleeping
eyes from the soft light of a brass lamp. She plucked Khemsa's arm
and made a quick gesture across her own throat. Khemsa lifted his
hand; then his expression changed and he drew away.
'I have eaten his salt,' he muttered. 'Besides, he can not interfere
with us.'
He led the girl through a door that opened on a winding stair. After
their soft tread had faded into silence, the man on the couch sat up.
Kerim Shah wiped the sweat from his face. A knife-thrust he did not
dread, but he feared Khemsa as a man fears a poisonous reptile.
'People who plot on roofs should remember to lower their voices,' he
muttered. 'But as Khemsa has turned against his masters, and as he
was my only contact between them, I can count on their aid no
longer. From now on I play the game in my own way.'
Rising to his feet he went quickly to a table, drew pen and
parchment from his girdle and scribbled a few succinct lines.

'To Khosru Khan, governor of Secunderam: the Cimmerian

Conan has carried the Devi Yasmina to the villages of the
Afghulis. It is an opportunity to get the Devi into our hands, as
the king has so long desired. Send three thousand horsemen at
once. I will meet them in the valley of Gurashah with native
guides.'

And he signed it with a name that was not in the least like Kerim
Shah.
Then from a golden cage he drew forth a carrier pigeon, to whose
leg he made fast the parchment, rolled into a tiny cylinder and
secured with gold wire. Then he went quickly to a casement and
tossed the bird into the night. It wavered on fluttering wings,
balanced, and was gone like a flitting shadow. Catching up helmet,
sword and cloak, Kerim Shah hurried out of the chamber and down
the winding stair.
The prison quarters of Peshkhauri were separated from the rest of
the city by a massive wall, in which was set a single iron-bound door
under an arch. Over the arch burned a lurid red cresset, and beside
the door squatted a warrior with spear and shield.
This warrior, leaning on his spear, and yawning from time to time,
started suddenly to his feet. He had not thought he had dozed, but a
man was standing before him, a man he had not heard approach.
The man wore a camel-hair robe and a green turban. In the
flickering light of the cresset his features were shadowy, but a pair of
lambent eyes shone surprizingly in the lurid glow.
'Who comes?' demanded the warrior, presenting his spear. 'Who are
you?'
The stranger did not seem perturbed, though the spear-point
touched his bosom. His eyes held the warrior's with strange
intensity.
'What are you obliged to do?' he asked, strangely.
'To guard the gate!' The warrior spoke thickly and mechanically; he
stood rigid as a statue, his eyes slowly glazing.
'You lie! You are obliged to obey me! You have looked into my eyes,
and your soul is no longer your own. Open that door!'
Stiffly, with the wooden features of an image, the guard wheeled
about, drew a great key from his girdle, turned it in the massive lock
and swung open the door. Then he stood at attention, his unseeing
stare straight ahead of him.
A woman glided from the shadows and laid an eager hand on the
mesmerist's arm.
'Bid him fetch us horses, Khemsa,' she whispered.
'No need of that,' answered the Rakhsha. Lifting his voice slightly he
spoke to the guardsman. 'I have no more use for you. Kill yourself!'
Like a man in a trance the warrior thrust the butt of his spear
against the base of the wall, and placed the keen head against his
body, just below the ribs. Then slowly, stolidly, he leaned against it
with all his weight, so that it transfixed his body and came out
between his shoulders. Sliding down the shaft he lay still, the spear
jutting above him its full length, like a horrible stalk growing out of
his back.
The girl stared down at him in morbid fascination, until Khemsa took
her arm and led her through the gate. Torches lighted a narrow
space between the outer wall and a lower inner one, in which were
arched doors at regular intervals. A warrior paced this enclosure,
and when the gate opened he came sauntering up, so secure in his
knowledge of the prison's strength that he was not suspicious until
Khemsa and the girl emerged from the archway. Then it was too
late. The Rakhsha did not waste time in hypnotism, though his
action savored of magic to the girl. The guard lowered his spear
threateningly, opening his mouth to shout an alarm that would bring
spearmen swarming out of the guardrooms at either end of the
alleyway. Khemsa flicked the spear aside with his left hand, as a
man might flick a straw, and his right flashed out and back, seeming
gently to caress the warrior's neck in passing. And the guard pitched
on his face without a sound, his head lolling on a broken neck.
Khemsa did not glance at him, but went straight to one of the
arched doors and placed his open hand against the heavy bronze
lock. With a rending shudder the portal buckled inward. As the girl
followed him through, she saw that the thick teakwood hung in
splinters, the bronze bolts were bent and twisted from their sockets,
and the great hinges broken and disjointed. A thousand-pound
battering-ram with forty men to swing it could have shattered the
barrier no more completely. Khemsa was drunk with freedom and
the exercise of his power, glorying in his might and flinging his
strength about as a young giant exercises his thews with
unnecessary vigor in the exultant pride of his prowess.
The broken door let them into a small courtyard, lit by a cresset.
Opposite the door was a wide grille of iron bars. A hairy hand was
visible, gripping one of these bars, and in the darkness behind them
glimmered the whites of eyes.
Khemsa stood silent for a space, gazing into the shadows from
which those glimmering eyes gave back his stare with burning
intensity. Then his hand went into his robe and came out again, and
from his opening fingers a shimmering feather of sparkling dust
sifted to the flags. Instantly a flare of green fire lighted the
enclosure. In the brief glare the forms of seven men, standing
motionless behind the bars, were limned in vivid detail; tall, hairy
men in ragged hill-men's garments. They did not speak, but in their
eyes blazed the fear of death, and their hairy fingers gripped the
bars.
The fire died out but the glow remained, a quivering ball of lambent
green that pulsed and shimmered on the flags before Khemsa's feet.
The wide gaze of the tribesmen was fixed upon it. It wavered,
elongated; it turned into a luminous greensmoke spiraling upward. It
twisted and writhed like a great shadowy serpent, then broadened
and billowed out in shining folds and whirls. It grew to a cloud
moving silently over the flags—straight toward the grille. The men
watched its coming with dilated eyes; the bars quivered with the
grip of their desperate fingers. Bearded lips parted but no sound
came forth. The green cloud rolled on the bars and blotted them
from sight; like a fog it oozed through the grille and hid the men
within. From the enveloping folds came a strangled gasp, as of a
man plunged suddenly under the surface of water. That was all.
Khemsa touched the girl's arm, as she stood with parted lips and
dilated eyes. Mechanically she turned away with him, looking back
over her shoulder. Already the mist was thinning; close to the bars
she saw a pair of sandalled feet, the toes turned upward—she
glimpsed the indistinct outlines of seven still, prostrate shapes.
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!

ebookname.com