Download the full version of the textbook now at textbookfull.

com

Complex Surveys Analysis of Categorical Data

1st Edition Parimal Mukhopadhyay (Auth.)

https://textbookfull.com/product/complex-surveys-
analysis-of-categorical-data-1st-edition-parimal-
mukhopadhyay-auth/

Explore and download more textbook at https://textbookfull.com

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

Categorical and Nonparametric Data Analysis E. Michael

Nussbaum

https://textbookfull.com/product/categorical-and-nonparametric-data-
analysis-e-michael-nussbaum/

textbookfull.com

A course in categorical data analysis First Edition

Leonard

https://textbookfull.com/product/a-course-in-categorical-data-
analysis-first-edition-leonard/

textbookfull.com

Analysis of Large and Complex Data 1st Edition Adalbert

F.X. Wilhelm

https://textbookfull.com/product/analysis-of-large-and-complex-
data-1st-edition-adalbert-f-x-wilhelm/

textbookfull.com

Magic for Liars A Novel 1st Edition Sarah Gailey

https://textbookfull.com/product/magic-for-liars-a-novel-1st-edition-
sarah-gailey/

textbookfull.com
Inventology how we dream up things that change the world
First Mariner Books Edition Kennedy

https://textbookfull.com/product/inventology-how-we-dream-up-things-
that-change-the-world-first-mariner-books-edition-kennedy/

textbookfull.com

Artistic Migration: Reframing Post-War Italian Art,

Architecture, and Design in Brazil 1st Edition Aline
Coelho Sanches
https://textbookfull.com/product/artistic-migration-reframing-post-
war-italian-art-architecture-and-design-in-brazil-1st-edition-aline-
coelho-sanches/
textbookfull.com

Riemannian Geometry and Geometric Analysis 7th Edition

Jürgen Jost (Auth.)

https://textbookfull.com/product/riemannian-geometry-and-geometric-
analysis-7th-edition-jurgen-jost-auth/

textbookfull.com

Endophytes Biology and Biotechnology Volume 1 1st Edition

Dinesh K. Maheshwari (Eds.)

https://textbookfull.com/product/endophytes-biology-and-biotechnology-
volume-1-1st-edition-dinesh-k-maheshwari-eds/

textbookfull.com

Handbook of Qualitative Health Research for Evidence Based

Practice 1st Edition Karin Olson

https://textbookfull.com/product/handbook-of-qualitative-health-
research-for-evidence-based-practice-1st-edition-karin-olson/

textbookfull.com
Political Party Membership in New Democracies: Electoral
Rules in Central and East Europe Alison F. Smith

https://textbookfull.com/product/political-party-membership-in-new-
democracies-electoral-rules-in-central-and-east-europe-alison-f-smith/

textbookfull.com
Parimal Mukhopadhyay

Complex
Surveys
Analysis of Categorical Data
Complex Surveys
Parimal Mukhopadhyay

Complex Surveys
Analysis of Categorical Data

123
Parimal Mukhopadhyay
Indian Statistical Institute
Kolkata, West Bengal
India

ISBN 978-981-10-0870-2 ISBN 978-981-10-0871-9 (eBook)

DOI 10.1007/978-981-10-0871-9

Library of Congress Control Number: 2016936288

© Springer Science+Business Media Singapore 2016

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer Science+Business Media Singapore Pte Ltd.
To
The Loving Memory of My Wife
Manju
Preface

Most of the data a statistician uses is categorical in nature. In the realms of bio-
medicine and social sciences, ecology and demography, voting pattern and mar-
keting, to name a few, categorical data dominate. They are the major raw materials
for analysis for testing different hypotheses relating to the populations which
generate such data. Such data are generally obtained from surveys carried under a
complex survey design, generally a stratified multistage design. In analysis of data
collected through sample surveys, standard statistical techniques are often routinely
employed.
We recall the well-recited phrase which we chanted in our undergraduate days:
Let x1,…, xn be a random sample of size n drawn from a population with probability
density function f(x) or probability mass function pM(x), etc. This means that the
sampled units whose observations are x1, x2,…, xn, are drawn by simple random
sampling with replacement (srswr). This also implies that observed variables x1,…,
xn are independently and identically distributed (IID). In fact, most of the results in
theoretical statistics, including those in usual analysis of categorical data, are based
on these assumptions.
However, survey populations are often complex with different cell probabilities
in different subgroups of the population, and this implies a situation different from
the IID setup. Longitudinal surveys—where sample subjects are observed over two
or more time points—typically lead to dependent observations over time.
Moreover, longitudinal surveys often have complex survey designs that involve
clustering which results in cross-sectional dependence among samples.
In view of these facts, it is, therefore, necessary to modify the usual tests of
goodness of fit, like Pearsonian chi-square, likelihood ratio and Wald statistic to
make them suitable for use in the context of data from complex surveys. A host of
ardent researchers have developed a number of such tests for this purpose over
more than last four decades.
There are already a myriad number of textbooks and research monographs on
analysis of categorical data. Then why is another book in the area required?
My humble answer is that all those treatise provide excellent description of the

vii
viii Preface

categorical data analysis under the classical setup (usual srswr or IID assumption),
but none addresses the problem when the data are obtained through complex
sample survey designs, which more often than not fail to satisfy the usual
assumptions. The present endeavor tries to fill in the gap in the area.
The idea of writing this book is, therefore, to make a review of some of the ideas
that have blown out in the field of analysis of categorical data from complex
surveys. In doing so, I have tried to systematically arrange the results and provide
relevant examples to illuminate the ideas. This research monograph is a review
of the works already done in the area and does not offer any new investigation. As
such I have unhesitatingly used a host of brilliant publications in this area. A brief
outline of different chapters is indicated as under:
(1) Chapter 1: Basic ideas of sampling; finite population; sampling design; esti-
mator; different sampling strategies; design-based method of making infer-
ence; superpopulation model; model-based inference
(2) Chapter 2: Effects of a true complex design on the variance of an estimator
with reference to a srswr design or an IID-model setup; design effects; mis-
specification effects; multivariate design effect; nonparametric variance
estimation
(3) Chapter 3: Review of classical models of categorial data; tests of hypotheses
for goodness of fit; log-linear model; logistic regression model
(4) Analysis of categorical data from complex surveys under full or saturated
models; different goodness-of-fit tests and their modifications
(5) Analysis of categorical data from complex surveys under log-linear model;
different goodness-of-fit tests and their modifications
(6) Analysis of categorical data from complex surveys under binomial and
polytomous logistic regression model; different goodness-of-fit tests and their
modifications
(7) Analysis of categorical data from complex surveys when misclassification
errors are present; different goodness-of-fit tests and their modifications
(8) Some procedures for obtaining approximate maximum likelihood estimators;
pseudo-likelihood approach for estimation of finite population parameters;
design-adjusted estimators; mixed model framework; principal component
analysis
(9) Appendix: Asymptotic properties of multinomial distribution; asymptotic
distribution of different goodness-of-fit tests; Neyman’s (1949) and Wald’s
(1943) procedures for testing general hypotheses relating to population
proportions
I gratefully acknowledge my indebtedness to the authorities of PHI Learning,
New Delhi, India, for kindly allowing me to use a part of my book, Theory and
Methods of Survey Sampling, 2nd ed., 2009, in Chap. 2 of the present book. I am
thankful to Mr. Shamin Ahmad, Senior Editor for Mathematical Sciences at
Preface ix

Springer, New Delhi, for his kind encouragement. The book was prepared at the
Indian Statistical Institute, Kolkata, to the authorities of which I acknowledge my
thankfulness. And last but not the least, I must acknowledge my indebtedness to my
family for their silent encouragement and support throughout this project.

January 2016 Parimal Mukhopadhyay

Contents

1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Fixed Population Model. . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Different Types of Sampling Designs . . . . . . . . . . . . . . . . . . . . . 8
1.4 The Estimators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.1 A Class of Estimators . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Model-Based Approach to Sampling . . . . . . . . . . . . . . . . . . . . . 17
1.5.1 Uses of Design and Model in Sampling . . . . . . . . . . . . . . 23
1.6 Plan of the Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2 The Design Effects and Misspecification Effects . . . . . . . . . . . . . . . . 27
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Effect of a Complex Design on the Variance of an Estimator . . . . 30
2.3 Effect of a Complex Design on Confidence Interval for θ . . . . . . . 37
2.4 Multivariate Design Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.5 Nonparametric Methods of Variance Estimation . . . . . . . . . . . . . 41
2.5.1 A Simple Method of Estimation of Variance
of a Linear Statistic . . . . . . . . . . . . . . . . . . . . . . . . .... 41
2.5.2 Linearization Method for Variance Estimation
of a Nonlinear Estimator . . . . . . . . . . . . . . . . . . . . . . . . 45
2.5.3 Random Group Method . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.5.4 Balanced Repeated Replications . . . . . . . . . . . . . . . . . . . 52
2.5.5 The Jackknife Procedures . . . . . . . . . . . . . . . . . . . . . . . . 56
2.5.6 The Jackknife Repeated Replication (JRR) . . . . . . . . . . . . 58
2.5.7 The Bootstrap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.6 Effect of Survey Design on Inference About
Covariance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 64
2.7 Exercises and Complements . . . . . . . . . . . . . . . . . . . . . . . .... 65

xi
xii Contents

3 Some Classical Models in Categorical Data Analysis . . . . . . . . . . . . 67

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Statistical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.2.1 Fitting Statistical Models . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.2 Large Sample Estimation Theory. . . . . . . . . . . . . . . . . . . 69
3.2.3 Asymptotic Properties of ML Estimates . . . . . . . . . . . . . . 70
3.2.4 Testing of Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.5 Transformation of the Central Limit Theorem . . . . . . . . . . 73
3.3 Distribution Theory for Count Data . . . . . . . . . . . . . . . . . . . . . . 73
3.3.1 Multinomial Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3.2 Poisson Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.3.3 The Multinomial-Poisson Connection. . . . . . . . . . . . . . . . 78
3.4 Goodness-of-Fit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.4.1 Likelihood Ratio Statistic . . . . . . . . . . . . . . . . . . . . . . . . 78
3.4.2 Pearson’s Goodness-of-Fit Statistic . . . . . . . . . . . . . . . . . 79
3.5 Binomial Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.5.1 Binomial Data and the Log-Likelihood Ratio . . . . . . . . . . 81
3.6 Log-Linear Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.6.1 Log-Linear Models for Two-Way Tables . . . . . . . . . . . . . 82
3.6.2 Log-Linear Models for Three-Way Tables . . . . . . . . . . . . 84
3.7 Logistic Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.7.1 The Logistic Regression Model. . . . . . . . . . . . . . . . . . . . 89
3.7.2 Fitting the Logistic Regression Model . . . . . . . . . . . . . . . 90
3.7.3 The Multiple Logistic Regression Model . . . . . . . . . . . . . 91
3.7.4 Fitting the Multiple Logistic Regression Model. . . . . . . . . 92
3.7.5 Polytomous Logistic Regression . . . . . . . . . . . . . . . . . . . 94
3.8 Fitting the Logistic Regression Models to Data
from Complex Surveys. . . . . . . . . . . . . . . . . . . . . . . ........ 95
4 Analysis of Categorical Data Under a Full Model . . . . . . . . . . . . . . 97
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.2 Tests of Goodness-of-Fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.2.1 Pearsonian Chi-Square Statistic . . . . . . . . . . . . . . . . . . . . 98
4.2.2 Design-Based Wald Statistic . . . . . . . . . . . . . . . . . . . . . . 99
4.2.3 Neyman’s (Multinomial Wald) Statistic . . . . . . . . . . . . . . 101
4.2.4 Log-Likelihood Ratio Statistic . . . . . . . . . . . . . . . . . . . . 101
2
4.2.5 Asymptotic Distribution of XW and XP2 . . . . . . . . . . . . . . . 101
4.2.6 Generalized Design Effect of π . . . . . . . . . . . . . . . . . . . . 107
4.2.7 Modification to XP2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.2.8 Fay’s Jackknifed Statistic . . . . . . . . . . . . . . . . . . . . . . . . 112
4.3 Testing for Homogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.3.1 A Simple Method for Binary Data
from Cluster Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.4 Effect of Survey Design on Classical Tests
of General Linear Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
Contents xiii

4.5 Tests of Independence in a Two-Way Table . . . . . . . . . . . . . . . . 123

4.6 Some Evaluation of Tests Under Cluster Sampling . . . . . . . . . . . 128
4.7 Exercises and Complements . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5 Analysis of Categorical Data Under Log-Linear Models . . . . . . . . . 135
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.2 Log-Linear Models in Contingency Tables . . . . . . . . . . . . . . . . . 136
5.3 Tests for Goodness of Fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.3.1 Other Standard Tests and Their First- and Second-Order
Corrections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.3.2 Fay’s Jackknifed Tests . . . . . . . . . . . . . . . . . . . . . . . . . . 143
5.4 Asymptotic Covariance Matrix of the Pseudo-MLE π ^ ....... . . 145
5.4.1 Residual Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.5 Brier’s Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.6 Nested Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.6.1 Pearsonian Chi-Square and the Likelihood
Ratio Statistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
5.6.2 A Wald Statistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.6.3 Modifications to Test Statistics . . . . . . . . . . . . . . . . . . . . 154
5.6.4 Effects of Survey Design on XP2 ð2j1Þ . . . . . . . . . . . . . . . . 155
6 Analysis of Categorical Data Under Logistic Regression Model . . . . 157
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.2 Binary Logistic Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.2.1 Pseudo-MLE of π . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.2.2 Asymptotic Covariance Matrix of the Estimators. . . . . . . . 160
6.2.3 Goodness-of-Fit Tests . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.2.4 Modifications of Tests . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.3 Nested Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.3.1 A Wald Statistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.3.2 Modifications to Tests . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.4 Choosing Appropriate Cell-Sample Sizes for Running Logistic
Regression Program in a Standard Computer Package . . . . . . . . . 169
6.5 Model in the Polytomous Case . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.6 Analysis Under Generalized Least Square Approach . . . . . . . . . . 172
6.7 Exercises and Complements . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
7 Analysis in the Presence of Classification Errors . . . . . . . . . . . . . . . 179
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
7.2 Tests for Goodness-of-Fit Under Misclassification . . . . . . . . . . . . 180
7.2.1 Methods for Considering Misclassification Under SRS. . . . 180
7.2.2 Methods for General Sampling Designs . . . . . . . . . . . . . . 182
7.2.3 A Model-Free Approach . . . . . . . . . . . . . . . . . . . . . . . . 183
xiv Contents

7.3 Tests of Independence Under Misclassification . . ......... . . . 185

7.3.1 Methods for Considering Misclassification Under SRS. . . . 186
7.3.2 Methods for Arbitrary Survey Designs . . . ......... . . . 186
7.4 Test of Homogeneity . . . . . . . . . . . . . . . . . . . . ......... . . . 188
7.5 Analysis Under Weighted Cluster Sample Design ......... . . . 192
8 Approximate MLE from Survey Data. . . . . . . . . . . . . . . . . . . . . . . 195
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
8.2 Exact MLE from Survey Data. . . . . . . . . . . . . . . . . . . . . . . . . . 196
8.2.1 Ignorable Sampling Designs . . . . . . . . . . . . . . . . . . . . . . 196
8.2.2 Exact MLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
8.3 MLE’s Derived from Weighted Distributions . . . . . . . . . . . . . . . 198
8.4 Design-Adjusted Maximum Likelihood Estimation. . . . . . . . . . . . 200
8.4.1 Design-Adjusted Regression Estimation
with Selectivity Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.5 The Pseudo-Likelihood Approach to MLE
from Complex Surveys. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
8.5.1 Analysis Based on Generalized Linear Model . . . . . . . . . . 209
8.5.2 Estimation for Linear Models . . . . . . . . . . . . . . . . . . . . . 212
8.6 A Mixed (Design-Model) Framework. . . . . . . . . . . . . . . . . . . . . 216
8.7 Effect of Survey Design on Multivariate Analysis
of Principal Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
8.7.1 Estimation of Principal Components . . . . . . . . . . . . . . . . 222

Appendix A: Asymptotic Properties of Multinomial Distribution . . . . . . 223

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
About the Author

Parimal Mukhopadhyay is a former professor of statistics at the Indian Statistical

Institute, Kolkata, India. He received his Ph.D. degree in statistics from the
University of Calcutta in 1977. He also served as a faculty member at the
University of Ife, Nigeria, Moi University, Kenya, University of South Pacific, Fiji
Islands and held visiting positions at the University of Montreal, University of
Windsor, Stockholm University and the University of Western Australia. He has
written more than 75 research papers in survey sampling, some co-authored and
eleven books on statistics. He was a member of the Institute of Mathematical
Statistics and elected member of the International Statistical Institute.

xv
Chapter 1
Preliminaries

Abstract This chapter reviews some basic concepts in problems of estimating a

finite population parameter through a sample survey, both from a design-based
approach and a model-based approach. After introducing the concepts of finite popu-
lation, sample, sampling design, estimator, and sampling strategy, this chapter makes
a classification of usual sampling designs and takes a cursory view of some estima-
tors. The concept of superpopulation model is introduced and model-based theory of
inference on finite population parameters and model parameters is looked into. The
role of superpopulation model vis-a-vis sampling design for making inference about
a finite population has been outlined. Finally, a plan of the book has been sketched.

Keywords Finite population · Sample · Sampling frame · Sampling design · Inclu-

sion probability · Sampling strategy · Horvitz–Thompson estimator · PPS sampling ·
Rao–Hartly–Cochran strategy · Generalized difference estimator · GREG · Multi-
stage sampling · Two-phase sampling · Self-weighting design · Superpopulation
model · Design-predictor pair · BLUP · Purposive sampling design

1.1 Introduction

The book has two foci: one is sample survey and the other is analysis of categorical
data. The book is primarily meant for sample survey statisticians, both theoreticians
and practitioners, but nevertheless is meant for data analysts also. As such, in this
chapter we shall make a brief review of basic notions in sample survey techniques,
while a cursory view of classical models for analysis of categorical data will be
postponed till the third chapter.
Sample survey, finite population sampling, or survey sampling is a method of
drawing inference about the characteristic of a finite population by observing only a
part of the population. Different statistical techniques have been developed to achieve
this end during the past few decades.
In what follows we review some basic results in problems of estimating a finite
population parameter (like, its total or mean, variance) through a sample survey. We
assume throughout most of this chapter that the finite population values are fixed

© Springer Science+Business Media Singapore 2016 1

P. Mukhopadhyay, Complex Surveys, DOI 10.1007/978-981-10-0871-9_1
2 1 Preliminaries

quantities and are not realizations of random variables. The concepts will be clear
subsequently.

1.2 The Fixed Population Model

First, we consider a few definitions.

Definition 1.2.1 A finite (survey) population P is a collection of a known number
N of identifiable units labeled 1, 2, . . . , N ; P = {1, . . . , N }, where i denotes the
physical unit labeled i. The integer N is called the size of the population.
The following types of populations, therefore, do not satisfy the requirements
of the above definition: batches of industrial products of identical specifications
(e.g., nails, screws) coming out from a production process, as one unit cannot be
distinguished from the other, i.e., the identifiability of the units is lost; population of
animals in a forest, population of fishes in a typical lake, as the population size is
unknown. Collection of households in a given area, factories in an industrial complex,
and agricultural fields in a village are examples of survey populations.
Let ‘y’ be a study variable having value yi on i (= 1, . . . , N ). As an example,
in an industrial survey ‘yi ’ may be the value added by manufacture by a factory
i. The quantity yi is assumed to be fixed and not random. Associated with P we
have, therefore, a vector of real numbers y = (y1 , . . . , y N ) . The vector y therefore
constitutes a parameter for the model of a survey population, y ∈ R N , the parameter
space. In a sample survey one is often interested in estimating a parameter function
N
θ(y), e.g., population total T (y) = T or Y (= i=1 yi ), population mean Ȳ or ȳ
 N
(= T /N ), population variance S = 2
i=1 (yi − ȳ) /(N − 1). This is done by
2

choosing a sample (a part of the population, defined below) from P in a suitable

manner and observing the values of y only for those units which are included in the
sample.
Definition 1.2.2 A sample is a part of the population, i.e., a collection of a suitable
number of units selected from the assembly of N units which constitute the survey
population P.
A sample may be selected in a draw-by-draw fashion by replacing a unit selected
at a draw to the original population before making the next draw. This is called
sampling with replacement (wr).
Also, a sample may be selected in a draw-by-draw fashion without replacing a
unit selected at a draw to the original population. This is called sampling without
replacement (wor).
A sample when selected by a with replacement (wr)-sampling procedure may be
written as a sequence:

S = i1 , . . . , in , 1 ≤ it ≤ N (1.2.1)

where i t denotes the label of the unit selected at the tth draw and is not necessarily
unequal to i t  for t = t  (= 1, . . . , n). For a without replacement (wor)-sampling
1.2 The Fixed Population Model 3

procedure, a sample may also written as a sequence S, with i t denoting the label of
the unit selected at the tth draw. Thus, here,

S = {i 1 , . . . , i n }, 1 ≤ i t ≤ N , i t = i t  for t = t  (= 1, . . . , N ) (1.2.2)

since, here, repetition of unit in S is not possible.

Arranging the units in the (sequence) sample S in an increasing (decreasing) order
of magnitudes of their labels and considering only the distinct units, a sample may
also be written as a set s. For a wr-sampling by n draws, a sample written as a set is,
therefore,

s = ( j1 , . . . , jν(S) ), 1 ≤ j1 < · · · < jν(S) ≤ N (1.2.3)

where ν(S) is the number of distinct units in S.

In a wor-sampling procedure, a sample of n-draws, written as a set, is

s = ( j1 , . . . , jn ), 1 ≤ j1 < · · · < jn ≤ N . (1.2.4)

As an example, if in a wr-sampling S = {2, 9, 4, 9}, the corresponding s is s =

(2, 4, 9) with ν(S) = 3, since there are only three distinct units. Similarly, if for
a wor sampling procedure, S = {4, 9, 1}, the corresponding s is s = (1, 4, 9) with
ν(S) = 3. Clearly, information on the order of selection and repetition of units which
is available in the (sequence) sample S is not available in the (set) sample s.
Definition 1.2.3 Number of distinct units in a sample is its effective sample size.
Number of draws in a sample is its nominal sample size. In (1.2.3), ν(S) is the effective
sample size, 1 ≤ ν(S) ≤ n. For a wor-sample of n-draws, ν(S) = ν(s) = n.
Note that a sample is a sequence or set of some units from the population and it
does not include their y-values. Thus in an agricultural survey if the firms with labels
7, 3, and 11 are included in the sample and if the yields of these field are 231, 38,
and 15 units, respectively, the sample comprised the firms labeled 7, 3, and 11 only;
their yield figures have nothing to do with the sample in this case.
Definition 1.2.4 The sample space is the collection of all possible samples and is
often denoted as S. Thus S = {S} or {s} accordingly as we are interested in S or s.
In a simple random sample with replacement (srswr) of n draws (defied in
Sect. 1.3), S consists of N n samples S. In a simple random sample without   replace-
ment (srsowr) (defined in Sect. 1.3), S consists of (N )n samples S and Nn samples
s where (a)b = a(a − 1) . . . (a − b + 1), a > b. If the samples s of all possible sizes
are considered in a wor-sampling procedure, there are 2 N samples in S.
Definition 1.2.5 Let A be the minimal σ -field over S and p be a probability measure
defined over A such that p(s) [or p(S)] denotes the probability of selecting s [or S],
satisfying
p(s)[ p(S)] ≥ 0
4 1 Preliminaries
 
 
p(s) p(S) = 1. (1.2.5)
S S

In estimating a finite population parameter θ(y) through sample surveys, one of the
main tasks of the survey statistician is to find a suitable p(s) or p(S). The collection
(S, p) is called a sampling design (s.d.), often denoted as D(S, p) or simply p. The
triplet (S, A, p) is the probability space for the model of the finite population.
The expected effective sample size of a s.d. p is, therefore,

 
N
E{ν(S)} = ν(S) p(S) = μP[ν(S) = μ] = ν. (1.2.6)
S∈S μ=1

We shall denote by ρν the class of all fixed effective size [FS(ν)] designs, i.e.,

ρν = { p : p(s) > 0 ⇐⇒ ν(S) = ν}.

A s.d. p is said to be noninformative if p(s)[ p(S)] does not depend on the y-values.
In this treatise, unless otherwise stated, we shall consider noninformative designs
only.
Basu (1958) and Basu and Ghosh (1967) proved that all the information relevant
to making inference about the population characteristic is contained within the set
sample s and the corresponding y-values. For this reason we shall mostly confine
ourselves to the set sample s.
The quantities  
πi = p(s), πi j = p(s)
s i s i, j


πi1 ,...,ik = p(s) (1.2.8)
s i 1 ,...,i k

are, respectively, the first order, second order, . . . , kth order inclusion probabilities
of units in a s.d. p. The following lemma states some relations among inclusion
probabilities and expected effective sample size of a s.d.

Lemma 1.2.1 For any s.d. p,

(i) πi + π j − 1 ≤ πi j ≤ min(πi , π j ),
N
(ii) πi = ν,
i=1
N
(iii) i= j=1 πi j = ν(ν − 1) + V (ν(S)).

If p ∈ ρν ,
N
(iv) πi j = (ν − 1)πi ,
 Nj (=i)=1
(v) i= j=1 πi j = ν(ν − 1).
1.2 The Fixed Population Model 5

Further, for any s.d. p,

θ(1 − θ) ≤ V {ν(S)} ≤ (N − ν)(ν − 1) (1.2.9)

where ν = [ν] + θ, 0 ≤ θ < 1, θ being the fractional part of ν.

The lower bound in (1.2.9) is attained by a s.d. p for which

P[ν(S) = [ν]] = 1 − θ andP[ν(S) = [ν] + 1] = θ.

Mukhopadhyay (1975) gave a s.d. with fixed nominal sample size n[ p(S) > 0 ⇒
n(S) = n∀S] such that V {ν(S)} = θ(1 − θ)/(n − [ν]), which is very close to the
lower bound in (1.2.9).
It is seen, therefore, that a s.d. gives the probability p(s) [or p(S)] of selecting a
sample s (or S), which, of course, belongs to the sample space. In general, it will be
a formidable task to select a sample using only the contents of a s.d., because one
has to enumerate all the possible samples in some order, calculate the cumulative
probabilities of selection, draw a random number in [0, 1], and select the sample
corresponding to the number so selected. It will be, however, of great advantages if
one knows the conditional probabilities of selection of units at different draws.
We shall denote by
pr (i) = probability of selecting i at the r th draw, r = 1, . . . , n;
pr (ir |i 1 , . . . , ir −1 ) = conditional probability of drawing ir at the r th draw given
that i 1 , . . . , ir −1 were drawn at the first, . . . , (r − 1)th draw, respectively;
p(i 1 , . . . , ir ) = the joint probability that (i 1 , . . . , ir ) are selected at the first,
. . . , r th draw, respectively.
All these probabilities must be nonnegative and we must have


N
pr (i) = 1, r = 1, . . . , n;
i=1


N
pr (ir |i 1 , . . . , ir −1 ) = 1.
ir =1

Definition 1.2.6 A sampling scheme (s.s.) gives the conditional probability of draw-
ing a unit at any particular draw given the results of the earlier draws.
A s.s., therefore, specifies the conditional probabilities pr (ir |i 1 , . . . , ir −1 ), i.e.,
it specifies the values p1 (i)(i = 1, . . . , N ), pr (ir |i 1 , . . . ir −1 ), ir = 1, . . . , N ; r =
2, . . . , n.
The following theorem shows that any sampling design can be attained through
a draw-by-draw mechanism.

Theorem 1.2.1 (Hanurav 1962; Mukhopadhyay 1972) For any given sampling
design, there exists at least one sampling scheme which realizes this design.
6 1 Preliminaries

Suppose now that the values x1 , . . . , x N of a closely related (related to y) aux-

iliary variable x on  Nunits 1, 2, . . . , N , respectively, are available. The quantity
Pi = xi / X, X = i=1 x k is called the size measure of unit i(= 1, . . . , N ) and
is often used in selection of samples. Thus in a survey of large-scale manufacturing
industry, say, jute industry, the number of workers in a factory may be considered
as a measure of size of the factory, on the assumption that a factory employing more
manpower will have larger value of output.
Before proceeding to take a cursory view of different types of sampling designs
we will now introduce some terms useful in this context.
Sampling frame: It is the list of all sampling units in the finite population from
which a sample is selected. Thus in a survey of households in a rural area, the list
of all the households in the area will constitute a frame for the survey. The frame
also includes any auxiliary information like measures of size, which is used for
special sampling techniques, such as stratification and probability proportional-
to-size sample selections, or for special estimation techniques, such as ratio or
regression estimates. All these techniques have been indicated subsequently.
However, a list of all the ultimate study units or ultimate sampling units may not
be always available. Thus in a household survey in an urban area where each
household is the ultimate sampling unit or ultimate study unit we do not generally
have a list of all such households. But we may have a list of census block units
within this area from which a sample of census blocks may be selected at the first
stage. This list is the frame for sampling at the first stage. Each census block again
may consist of several wards. For each selected census block one may prepare a list
of such wards and select samples of wards. These lists are then sampling frames
for sampling at the second stage. Multistage sampling has been discussed in the
next section. Sarndal et al. (1992), among others, have investigated the relationship
between the sampling frame and population.
Analytic and Descriptive Surveys: Descriptive uses of surveys are directed at the
estimation of summary measures of the population such as means, totals, and
frequencies. Such surveys are generally of prime importance to the Government
departments which need an accurate picture of the population in terms of its loca-
tion, personal characteristics, and associated circumstances. The analytic surveys
are more concerned with identifying and understanding the causal mechanisms
which underlie the picture which the descriptive statistics portray and are gener-
ally of interest to research organizations in the area. Naturally, the estimation of
different superpopulation parameters, such as regression coefficients, is of prime
interest in such surveys.
For descriptive uses the objective of the survey is essentially fixed. Target parame-
ters, such as the total and ratio, are the objectives determined even before the data
are collected or analyzed. For analytic uses, such as studying different parameters
of the model used to describe the population, the parameters of interest are not
generally fixed in advance and evolve through an adaptive process as the analysis
progresses. Thus for analytic purposes the process is an evolutionary one where the
final parameters to be estimated and the estimation procedures to be employed are
1.2 The Fixed Population Model 7

chosen in the light of the superpopulation model used to describe the population.
Use of superpopulation model in sampling has been indicated in Sect. 1.5.
Strata: Sometimes, it may be necessary or desirable to divide the population into
several subpopulations or strata to estimate population parameters like population
mean and population total through a sample survey. The necessity of stratification
is often dictated by administrative requirements or convenience. For a statewide
survey, for instance, it is often convenient to draw samples independently from
each county and carry out survey operations for each county separately. In practice,
the population often consists of heterogeneous units (with respect to the character
under study). It is known that by stratifying the population such that the units
which are approximately homogeneous (with respect to ‘y  ), a better estimator of
population total, mean, etc. can be achieved.
We shall often denote by yhi the value  of y on the ith unit in the hth stra-
tum (i = 1, . . . , Nh ; h = 1, . . . , H ), h Nh = N , the population size; Ȳh =
 Nh  Nh
i=1 Yhi /Nh , Sh = i=1 (Yhi − Ȳh ) /(Nh − 1), stratum population mean and
2 2

variance, respectively;
H Wh = Nh /N , stratum proportion. The population mean is
then Ȳ = h=1 Wh Ȳh .
Cluster: Sometimes, it is not possible to have a list of all the units of study in the
population so that drawing a sample of such study units is not possible. However,
a list of some bigger units each consisting of several smaller units (study units)
may be available from which a sample may be drawn. Thus, for instance, in a
socioeconomic survey, our main interest often lies in the households (which are
now study units or elementary units or units of our ultimate interest). However, a
list of households is not generally available, whereas a list of residential houses
each accommodating a number of households should be available with appropriate
authorities. In such cases, samples of houses may be selected and all the households
in the sampled houses may be studied. Here, a house is a ‘cluster.’ A cluster consists
of a number of ultimate units or study units. Obviously, the clusters may be of
varying sizes. Generally, all the study units in a cluster are of the same or similar
character. In cluster sampling a sample of clusters is selected by some sampling
procedure and data are collected from all the elementary units belonging to the
selected clusters.
Domain: A domain is a part of the population. In a statewide survey, a district
may be considered as a domain; in the survey of a city a group of blocks may
form a domain, etc. After sampling has been done from the population as a whole
and the field survey has been completed, one may be interested in estimating
the mean or total relating to some part of the population. For instance, after a
survey of industries has been completed, one may be interested in estimating the
characteristic of the units manufacturing cycle tires and tubes. These units in the
population will then form a domain. Clearly, sample size in a domain will be a
random variable. Again, the domain size may or may not be known.
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
8 1 Preliminaries

1.3 Different Types of Sampling Designs

The following types of sampling designs are generally used.

(a) Simple random sampling with replacement (srswr): Under this scheme units
are selected one by one at random in n (a preassigned number) draws from
the list of all available units such that a unit once selected is returned to the
population before the next draw. As stated before, the sample space here consists
of N n sequences S{i 1 , . . . , i n } and the probability of selecting any such sequence
(sample) is 1/N n .
(b) Simple random sampling without replacement (srswor): Here units are selected
in n draws at random from the list of all available units such that a unit once
selected is removed from the population before the next draw. Here   again, as
stated before the sample space consists of (N )n sequences S and Nn sets s and
the s.d. design allots to each of them equal probability of selection.
(c) Probability proportional to size with replacement (ppswr) sampling: a unit i is
selected with probability pi at the r th draw and a unit once selected is returned
to the population before the next draw (i = 1, . . . , N ; r = 1, 2, . . . , n). The
quantity pi is a measure of size of the unit i. This s.d. is a generalization of srswr
s.d. where pi = 1/N ∀ i.
(d) Probability proportional to size without replacement (ppswor): a unit i is selected
at the r th draw with probability proportional to its normed measure of size and
a unit once selected is removed from the population before the next draw. Here,

p1 (i 1 ) = pi1

pr (ir |i 1 , . . . , ir −1 ) = pir /(1 − pi1 − · · · − pir −1 ), r = 2, . . . , n.

For n = 2, for this scheme,


pi
πi = pi 1 + A − ,
1 − pi

1 1 
N
pk
πi j = pi p j + , where A = .
1 − pi 1 − pj k=1
1 − pk

This sampling scheme is also known as ‘successive sampling.’ The correspond-

ing sampling design may also be attained by an inverse sampling procedure
where units are drawn by ppswr, until for the first time n distinct units occur.
The n distinct units each taken only once constitute the sample.
(e) Rejective sampling: n draws are made with ppswr; if all the units turn out to
be distinct, the selected sequence constitutes the sample; otherwise, the whole
selection is rejected and fresh draws made.
1.3 Different Types of Sampling Designs 9

(f) Unequal probability without replacement (upwor) sampling: A unit i is selected

at the r th draw with probability proportional to pi(r ) and a unit once selected is
removed from the population. Here

p1 (i) = pi(1)

pi(rr )
pr (ir |i 1 , . . . , ir −1 ) = , r = 2, . . . , n. (1.3.1)
1 − pi(r1 ) − pi(r2 ) − · · · pi(rr −1)

The quantities { pi(r ) } are generally functions of pi and p-values of the units
already selected. In particular, ppswor sampling scheme described in item (d)
above is a special case of this scheme, where pi(r ) = pi , r = 1, . . . , n. The
sampling design may also be attained by a inverse sampling procedure where
units are drawn wr, with probability pi(r ) at the r th draw, until for the first time
n distinct units occur. The n distinct units each taken only once constitute the
sample.
(g) Generalized rejective sampling: Draws are made wr and with probability { pi(r ) }
at the r th draw. If all the units turn out distinct, the solution is taken as a sample;
otherwise, the whole sample is rejected and fresh draws are made. The scheme
reduces to the scheme at (e) above, if pi(r ) = pi ∀i.
(h) Systematic sampling with varying probability (including unequal probability).
(k) Sampling from groups: The population is divided into L groups either at random
or following some suitable procedure and a sample of size n h is drawn from the
hth group using any of the  above-mentioned sampling designs such that the
desired sample size n = h n h is attained. Examples are stratified sampling
procedure and Rao–Hartley–Cochran (1962) (RHC) sampling procedure. Thus
in stratified random sampling the population is divided into H strata of sizes
N1 , . . . , N H and a sample of size n h is selected at random from the hth stratum
(h = 1, . . . , H ). The quantities n h and n = h n h are suitably determined.
RHC procedure has been discussed in the next section.

Based on the above methods, there are many unistage or multistage stratified
sampling procedures. In a multistage procedure sampling is carried out in many
stages. Units in a two-stage population consist of N first-stage units (fsu’s) of sizes
M1 , . . . , M N , with the bth second stage unit (ssu) in the ath fsu being denoted ab
for a = 1, . . . , N ; b = 1, . . . , Ma , with its associated y values being denoted yab .
For a three-stage population the cth third stage unit (tsu) in the bth ssu in the ath
fsu is labeled abc for a = 1, . . . , N ; b = 1, . . . , Ma ; c = 1, . . . , K ab . In a three-
stage sampling a sample of n fsu’s is selected out of N fsu’s and from each of the ath
selected fsu’s, a sample of m a ssu’s is selected out of Ma fsu’s in the selected fsu (a =
1, . . . , n). At the third stage from each of the selected ab ssu’s, containing K ab tsu’s
a sample of kab tsu’s is selected (a = 1, . . . , n; b = 1, . . . , m a ; c = 1, . . . , kab ). The
associated y-value is denoted as yabc , c = 1, . . . , kab ; b = 1, . . . , m a ; a = 1, . . . , n.
10 1 Preliminaries

The sampling procedure at each stage may be srswr, srswor, ppswr, upwor, sys-
tematic sampling, Rao–Hartley–Cochran sampling or any other suitable sampling
procedure. The process may be continued to any number of stages. Moreover, the
population may be initially divided into a number H of well-defined strata before
undertaking the stage-wise sampling procedures. For stratified multistage population
the label h is added to the above notation (h = 1, . . . , H ). Thus here the unit in the
hth stratum, ath fsu, bth ssu, and cth tsu is labeled habc and the associated y value
as yhabc .
As is evident, samples for all the sampling designs may be selected by a whole
sample procedure or mass-draw procedure in which a sample s is selected with
probability p(s).
A F.S.(n)-s.d. with πi proportional to pi = xi / X , where  N xi is the value of a
closely related (to y) auxiliary variable on unit i and X = k=1 xk , is often used for
estimating a population total. This is, because an important estimator, the Horvitz–
Thompson estimator (HTE) has very small variance if yi is nearly proportional to pi .
(This fact will be clear in the next section.) Such a design is called a πps design or
I P P S (inclusion probability proportional to size) design. Since πi ≤ 1, it is required
that xi ≤ X/n for such a design.
Many (exceeding seventy) unequal probabilities without replacement sampling
designs have been suggested in the literature, mostly for use along with the HTE.
Many of these designs attain the π ps property exactly, some approximately. For
some of these designs, such as the one arising out of Poisson sampling, sample
size is a variable. Again, some of these sampling designs are sequential in nature
(e.g., Chao 1982; Sunter 1977). Mukhopadhyay (1972), Sinha (1973), and Herzel
(1986) considered the problem of realizing a sampling design with preassigned sets
of inclusion probabilities of first two orders.
Again, in a sample survey, all the possible samples are not generally equally
preferable from the point of view of practical advantages. In agricultural surveys,
for example, the investigators tend to avoid grids which are located further away
from the cell camps, which are located in marshy land, inaccessible places, etc.
In such cases, the sampler would like to use only a fraction of the totality of all
possible samples, allotting only a very mall probability to the non-preferred units.
Such sampling designs are called Controlled Sampling Designs.
Chakraborty (1963) used a balanced incomplete block (BIB) design to obtain
a controlled sampling design replacing a srswor design. For unequal probability
sampling BIB designs and t designs have been considered by several authors (e.g.,
Srivastava and Saleh 1985; Rao and Nigam 1990; Mukhopadhyay and Vijayan 1996).
For a review of different unequal probability sampling designs the reader may
refer to Brewer and Hanif (1983), Chaudhuri and Vos (1988), Mukhopadhyay (1996,
1998b), among others.
1.4 The Estimators 11

1.4 The Estimators

After the sample has been selected, the statistician collects data from the field. Here,
again the data may be collected with respect to a sequence sample or set sample.

Definition 1.4.1 Data collected through a sequence sample S are

d  = {(k, yk ), k ∈ S}. (1.4.1)

For the set sample data are

d = {(k, yk ), k ∈ s}. (1.4.2)

It is known that data given in (1.4.2) are sufficient for making inference about θ,
whether the sample is a sequence S or a set s (Basu and Ghosh 1967). Data are
said to be unlabeled if after the collection of data its label part is ignored. Unlabeled
data may be represented by a sequence or a set of the observed values without any
reference to the labels.
It may not be possible, however, to collect the data from the sampled units cor-
rectly and completely. If the information is collected from a human population, the
respondent may not be ‘at home’ during the time of collection of data or may refuse
to answer or may give incorrect information, e.g., in stating income, age, etc. The
investigators in the field may also fail to register correct information due to their own
lapses.
We assume throughout that our data are free from such types of errors due to
non-response and errors of measurement and it is possible to collect the information
correctly and completely.

Definition 1.4.2 An estimator e = e(s, y) or e(S, y) is a function defined on S × R N

such that for a given (s, y) or (S, y), its value depends on y only through those i for
which i ∈ s (or S).
Clearly, the value of e in a sample survey does not depend on the units not included
in the sample.
An estimator e is unbiased for T with respect to a sampling design p if

E p (e(s, y)) = T ∀y ∈ R N (1.4.3)

i.e., 
e(s, y) p(s) = T ∀y ∈ R N
s∈S

where E p , V p denote, respectively, expectation and variance with respect to the

s.d. p. We shall often omit the suffix p when it is clear otherwise. This unbiasedness
will sometimes be referred to as p-unbiasedness.
12 1 Preliminaries

The mean square (MSE) of e around T with respect to a s.d. p is

M(e) = E(e − T )2
(1.4.4)
= V (e) + (B(e))2

where B(e) denotes the design bias, E(e) − T . If e is unbiased for T, B(e) vanishes
and (1.4.4) gives the variance of e, V (e).

Definition 1.4.3 A combination ( p, e) is called a sampling strategy, often denoted as

H ( p, e). This is unbiased for T if (1.4.3) holds and then its variance is V {H ( p, e)} =
E(e − T )2 .
A unbiased sampling strategy H ( p, e) is said to be better than another unbiased
sampling strategy H  ( p  , e ) in the sense of having smaller variance, if

V {H ( p, e)} ≤ V {H  ( p, e )} ∀y ∈ R N (1.4.5)

with strict inequality for at least one y.

If the s.d. p is kept fixed, an unbiased estimator e is said to be better than another
unbiased estimator e in the sense of having smaller variance, if

V p (e) ≤ V p (e ) ∀ y ∈ R N (1.4.6)

with strict inequality holding for at least one y.

We shall now consider different types of estimators for ȳ, when the s.d. is srswor,
based on n draws.

(1) Mean per unit estimator: Ȳˆ = ȳs = i∈s yi /n
 N Var ( ȳs ) =2 (1 − f )S /n
2
Variance:
S 2 = i=1 (yi − Ȳ ) /(N − 1)
N
Ȳ = i=1 yi /N , f = N /n
(2) Ratio estimator: ȳˆ R = ( ȳs /x̄s ) X̄
Mean square error: MSE ( ȳˆ R ) ≈ ( 1−n f )[S y2 + Sx2 − 2RS yx ],
N N
R = Y/ X, S y2 = S 2 , Sx2 = i=1 (xi − X̄ )2 /(N − 1), X = i=1 xi ,
N
X̄ = X/N Sx y = i=1 (yi − Ȳ )(xi − X̄ ).
(3) Difference estimator: ȳˆ D = ȳs + d( X̄ − x̄s ), where d is a known constant.
Variance : Var ( ȳˆ D ) = ( 1−n f )(S y2 + d 2 Sx2 − 2d Sx y ).
(4) Regression
 estimator: ȳˆlr = ȳs+ b( X̄ − x̄s )
b = i∈s (yi − ȳs )(xi − x̄s )/ i∈s (xi − x̄s )2 .
Mean square error: MSE ( ȳˆlr ) ≈ ( 1−n f )S y2 (1 − ρ2 )
where ρ is the correlation coefficient between x and y.
(5) Meanof the ratios estimator: ȳˆ M R = X̄ r̄
r̄ = i∈s ri /n
Random documents with unrelated
content Scribd suggests to you:
 When Dr. Olson had done likewise Chanler looked long and
lovingly at Miss Baldwin.
“Betty,” he said, as if rousing himself with an effort.
“Yes, George.”
“Betty, don’t you think you were an awful fool to come on a crazy
trip like this?”
She smiled as if humoring him.
“Why do you say that, George?”
“Suppose folks should hear about it?”
“What then?”
“Betty—you—all alone on a yacht with me. What’ll folks think if
they know?”
“They do know,” she said. “I told my folks and friends where I
was going.”
“Yes, but you told them my sister was on board.”
“Certainly—as you told me.”
“Oh, don’t rub it in, Betty. That’s past. But what do you think
people will think when they know she wasn’t on board, and that you
came ’way up here alone to join me?”
She looked at him steadily. I half rose to leave, but a glance from
her eyes told me to remain. It was not a pleasant scene. I stared at
my napkin.
“You see, Betty,” he continued, leaning loosely across the table,
“that’s what it will look like. Won’t it, Gardy?”
I did not reply.
“What will it look like, George?” she asked evenly.
“Like you were chasing me.”
She laughed, and her laughter was like a song-burst of
wholesome young life in the atmosphere of Chanler’s drink-drugged
maundering.
“Well, George, isn’t that what I am doing?”
“People will talk, Betty,” he persisted. “It’s a bad situation—for
you. I—I’m sorry I got you to come here—no, hang it! I’m not. But I
am worrying about your reputation, Betty.”
 “I think I can take care of my reputation, George,” she said
quietly.
“Let me take care of it, Betty!” he cried hoarsely, taking her hand.
“Please, George,” she said, smiling, as she rose.
“Betty!” He clung to her hand.
With swift, confident strength she drew her hand free, lifting him
slightly from his chair in doing so.
“You’ll excuse me now, won’t you?” she said, and went to her
room.
Chanler flung himself back in his chair, laughing harshly.
“Did you see that—did you see it, Gardy?” he said, as he pressed
the bell. “She doesn’t care if I do own this yacht. I’m nothing to her.
Oh, what a rotten trip this is going to be!”
“Chanler,” I said, “sit still for a minute and listen. You have got to
pull yourself together. You have got to straighten out this mess. You
have got to show Miss Baldwin that you are the man she is hoping
to find in you. Buck up, man! Her hopes are pinned on you. She
cares. Do you think she would have come this far if she didn’t care?
She has done her share; she’s here. Now, for her sake, do your
share. Pull yourself together and be the man she has been hoping all
this time she would find you.”
“Hooray!” he whispered mockingly. “Go on, Gardy; you’re the boy
who can say things. King’s peg,” he said to the steward who had
come in.
“Wait!” I said. The man stopped. “Chanler, you’ve been overdoing
it. You’re not yourself. You’ve done things that aren’t done; you’ve
got to sober up and straighten them out.”
“Got to!”
“Yes; as a gentleman you’ve got to. Miss Baldwin’s happiness—
perhaps her whole life’s happiness—depends on your being a
gentleman from now on. For God’s sake man! Isn’t it worth sobering
up to win a prize like that?”
“Oh, leave me alone, Gardy,” he growled. “Don’t you think I know
what I’m doing? It doesn’t make any difference what I do now. I’ve
lost her. She wouldn’t have me no matter what I did now. I know it.
Knew it five minutes after she came on board. Saw it in her eyes.
Felt it. My hold on her’s slipped—just like that. Gone—forever. No
use trying. King’s peg,” he repeated, “and hurry.”
I sat silent, rage and disgust choking me, while the man brought
in that terrible mixture of champagne and brandy in equal parts.
Chanler drank it in gulps.
“Have some, Gardy? No? That’s right. Some men shouldn’t touch
rum; you’re one of them. ’Cause why? ’Cause you’ve got a
conscience. Rot, rot, rot! Got to straighten up, have I, Gardy? ‘Got
to’ are words that weren’t made for me, my boy.”
“For God’s sake! Chanler, drop that sort of talk!” I cried, springing
to my feet. “If you knew what a sickening parody you are on the
gentleman you were at home, you wouldn’t put on airs.”
“Not to me, Gardy, not to me can you utter such contemptuous
words,” he said harshly.
“You be ——, you and your big talk!” I exploded. “Do you think
you’re entitled to any respect? Do you think I or any one else on
board cares who you are at present? Do you think your money is still
a power? Well, it’s not. It ceased to be this morning. Brack and the
crew—Brack especially—there’s the power aboard this yacht. And
you’re disgracing yourself and your class before them all.
“First you lie by wireless to get Miss Baldwin on board, and now
you’re taking the easiest way, keeping drunk, because you’re not
man enough to face the situation sober—not man enough to make
things right for the girl who came here trustfully depending on you.
Think of it, Chanler; think who you are—of your family. Have one
more try at decency, at least. Chuck away that poison in your hand
and let me call Dr. Olson and get you straightened up.”
He raised the large glass to his lips and drank the peg down
without a falter.
“Gardy,” he said, setting the glass down, “you’re fired.”
I laughed.
“I like you, Gardy; you’re a dear old fellow,” he continued, “but
you mustn’t presume on our friendship and talk to me like that. I’ve
got to let you out.”
“And I suppose I’m to pack my things and go?” said I. “Oh, come,
Chanler; wake up. Try to see things with sane eyes. I don’t care
whether I’m fired or whether we remain friends. We’re all on the
same plane for the present; you, Miss Baldwin, myself, we’re in the
hands of Captain Brack and the crew.”
He shuddered nervously.
“Don’t say such things, Gardy; I forbid them in my hearing.”
“You’re afraid to hear them, you mean.”
“Afraid or not, it makes no difference. They annoy me and I won’t
be annoyed. I won’t, you hear. Been annoyed enough on this trip.
Here I was waiting for Betty’s coming. Felt sure she’d have me if I
got her away alone, just herself and me. She comes, looks around. I
look in her eyes and bang! I see she won’t have me. Plain as print.
Whole trip useless. It’s a rotten world!”
“You’re giving up without a struggle, Chanler?”
“No use, my boy. I don’t like struggling, anyhow.”
“But, Miss Baldwin is, at least your guest, on board your yacht.
The yacht is in the hands of Brack and the crew. Haven’t you
thought that this situation might develop into one that may be
unpleasant and even unsafe for Miss Baldwin?”
“I have,” he said, signaling for another peg. “And I wish I was
back home in the big leather chair at the club, looking out on Fifth
Avenue.” He waved his hand drunkenly toward me. “I entrust—
entrust Miss Beatrice Baldwin—safety, pleasure, honor, rep’tation to
you, Gardy. Ha! There’s a bright little idea. I hire you again, Gardy.
New job. You—you see Betty safe and sound back to her folks.”
That hour marked the beginning of Chanler’s eclipse. At dinner-
time Simmons reported him indisposed. During the next three days
he left his room but seldom. He had but one desire now: to
eliminate himself as a responsible factor in the storm of events
about to break upon the Wanderer and its people.
 XVIII

Captain Brack was sitting in Chanler’s chair when we went in to

dinner that evening and Miss Baldwin’s place was beside him. Dr.
Olson and myself—neither Riordan nor Wilson had appeared—sat
opposite.
Brack was dressed with the care of a captain of a popular trans-
Atlantic liner, and his attitude toward Miss Baldwin was solely that of
a captain solicitous for his passenger’s comfort and pleasure. The
yacht might have been the Mauretania, our little party the dinner
crowd of the liner’s first saloon. Brack’s personality, polished and
radiant for the time being, his flashing conversation, filled and
illumined the room. It was difficult not to forget young Larson as one
sat beneath his spell.
“An apology is necessary, Miss Baldwin, for my absence from
luncheon,” he said. “It is not etiquette to fail to welcome a
passenger to her first meal on board. It was necessary, however,
that I stay on the bridge until I was sure that the Wanderer had
reached her limit of speed and that we were holding true on our
course. I have stolen thirty minutes from that duty this evening to
fulfil my social obligation as captain.”
“Then we are in a hurry, Captain Brack?” she asked.
His eyes were upon her—those eyes with their compelling power
—and her manner was subdued.
“The crew is in a desperate hurry, Miss Baldwin,” he said with one
of his flashing smiles. “Men are always in a hurry when they hear of
gold. And, really—” he bowed to her deferentially—“we have much
to thank you for, Miss Baldwin, for relieving a tense situation this
morning. I do not mean that there was the slightest danger of any
trouble. No, no! But the situation was a trifle uncomfortable when
you appeared and voted that we go hunting for gold instead of
bones.” He laughed softly. “I have wondered why you did that, Miss
Baldwin; is it presumptuous to ask?”
Miss Baldwin toyed with her spoon.
“I thought that this—going gold-hunting—was so much more
alive.”
“Good!” he said earnestly. “That is why I voted for it, too. To be
alive while we are living—that is more important than to unearth old
skeletons. Isn’t that your idea, Miss Baldwin?”
“Yes,” she said with a strange smile.
“And to be alive means to live in the open, free and untrapped.”
She looked up at him, and by her expression I knew that she saw
only his eyes.
“You don’t look as if you would be contented indoors, captain,”
she said with a little laugh.
“Are you?” he said, and looked straight at her.
She smiled in puzzled fashion without replying.
“No, you are not,” he answered for her. “For you are very, very
much alive, and so must naturally have longings for the free life,
which means life outdoors. Am I not right?”
“Yes.”
“Life—we can make it a free, glorious thing, or a gray, trapped
affair, just as we choose. It is all a matter of courage. There is still
much room in the world. It is not crowded except in spots. If we
choose to remain in one of those crowded spots, or rather, if we are
afraid to leave them, we must, of necessity, become one of the gray,
trapped crowd, existing through a certain span of years without ever
knowing what it is to be truly alive. But in the great open spaces
people live—they are alive. They are natural, they are hand-in-hand
with Nature, and Nature gives them more reward for living than
does what man calls civilization.
“As one who has lived under both conditions, Miss Baldwin, I
assure you that it is only in the uncrowded spaces that man may get
close enough to the root of Life to experience the sensation of
immortality. Haven’t you felt something like that yourself?”
 “Yes,” she said again, and her eyes were puzzled and full of
wonder.
“You will learn,” he said, nodding his head gravely. “You are one of
those who will learn quickly the message that the open has for you.
You are free-born. You would not be here unless the call to freedom
had come to you. Isn’t that so?”
“I—I have always longed for an experience like this. How did you
know?”
“It is written upon you as plain as print; you are finding your true
sphere. Tell me truthfully: do you not at this moment feel stirred as
you never did before in your life?”
She looked up at him quickly; it seemed as if he had frightened
her.
“How could you know that?” she faltered.
He smiled, leaning toward her, his eyes holding hers.
“That and many more things you will learn, Miss Baldwin,” he said
impressively. “You are beginning a new life. The new impulses you
feel are the commands of your true spirit, stricken free of the bonds
of civilization. Obey them. Remember, they are your true self; there
can be for you no realization of the full possibilities of life save along
the way they lead you. There is hidden country in all of us, and until
we explore it we don’t know what it is to live.”
He sat back in his chair, smiling, satisfied.
“And now you must excuse me; my thirty minutes are up and I
have promised Riordan thirty minutes to dine.” As he bowed and
rose his glance went across the table to me. “Now, Mr. Pitt, I will
wager, never has felt a call to be free—to explore any hidden
country.”
I did not reply.
“No, Mr. Pitt is not one of us. But, Miss Baldwin,” he concluded,
bending over her as he passed out, “you are. Your true life is about
to begin.”
And she followed him with her eyes as he left the room, though
there was that in her expression which suggested that she did so
unwillingly.
 “Ah!”
The faintest exclamation of relief escaped her lips as the captain
disappeared. She sank back in her chair as if suddenly released. She
looked around; our eyes met. She excused herself in a dazed sort of
fashion and went to her room.

Hours afterward I was pacing the deck. It was another pitch-dark

night, and to one fresh from the glare of New York, the darkness
was well-nigh appalling. The Wanderer’s searchlight seemed only a
thin knife-gash, parting the darkness before us. On either side of its
beam the blackness of night stood like a wall. There were no stars to
be seen above. East, north, south and west, naught but the dead
night; below, only the hiss of unseen waters through which we were
rushing toward—what?
I shuffled to and fro on the deck, caring neither where nor how I
was going. The scene between Brack and Miss Baldwin at the
dinner-table repeated itself again and again, each time with a new,
sinister significance. I know what power lay within Brack’s eyes. Had
they not roused me and thrilled me and made me fighting mad,
which was exactly what Brack, in idle sport wished to do? What
would be the effect of his will, gleaming through his glances, on a
woman, on a young, inexperienced girl like Miss Baldwin? For after
all, she was nothing but an inexperienced girl. Yes, I told myself, she
was so inexperienced, so ignorant, through the sheltered life she
had lived, that she did not know enough to recognize a distressing
situation when she met it. She was brave because she didn’t have
sense enough to be cautious.
“Mr. Pitt,” called a voice softly, “is that you?”
I swung around. I was near a cabin porthole and by its light I
made out Miss Baldwin coming toward me.
“I’m glad,” she said. “Don’t stop, please; let us walk.
“I came out,” she continued, as we fell into step, “because I didn’t
like to be alone.”
“Why not?”
 “I don’t know. I seemed lonesome. It was nice to come out here
and find you.”
I made no response, and our walk was silent for a long time.
“I wanted to speak to you about something,” she said at last,
“about Captain Brack.”
“Yes?”
She hesitated.
“Is—is he as wonderful as he seems?”
“Captain Brack is a remarkable man,” I replied.
“I thought he was wonderful when he was speaking,” she said
falteringly. “But when he was gone I—it seemed different.”
“How different?”
“I don’t know just. I loved to listen while he was talking. But after
he’d gone I felt relieved. It frightened me a little. That’s why I came
out. What do you know about him?”
I was at loss for a reply. To tell her what I knew of Brack, of my
first sight of him in the Seattle saloon, of what I had learned aboard
the Wanderer, would serve to alarm her in an uncomfortable
manner.
“Chanler selected him as his captain,” I said.
She gave an impatient toss to her shoulders as we walked on.
“Oh, that doesn’t mean anything. What sort of a man is he?”
“Very strong.”
“I know that.”
“Very capable.”
“Yes.”
“And entirely unscrupulous.”
She nodded her head, not in the least surprised.
“I thought so,” she said.
There was a moment of silence. We heard the murmur of waters
against our bows.
“He’s something like that,” she said, pointing out over the dark
sea. “A blind, remorseless force; isn’t he?”
“But more subtle.”
 “Oh! Is he?”
“As subtle as he is strong.”
She gave a little gasp, as if she had caught herself in an error.
“I didn’t know that. I didn’t realize—I must be going in. You’ll
excuse me. Good night, Mr. Pitt. Pleasant dreams.”
Pleasant dreams! It was past one in the morning before I ceased
my troubled pacing of the Wanderer’s promenade, and such sleep as
weariness finally brought to me was beset by a jumble of
nightmares, dominated by Brack’s eyes and smile.
 XIX

After breakfast next morning I went to see Chanler. He was sitting

up in bed, and he had changed greatly overnight. His face was
puffed and gray-looking, and the swollen eyelids were parted only
enough to disclose a slit of blood-shot eyes. Dr. Olson was with him,
whisky-glass in hand, but he was watching Chanler shrewdly.
“I’ve got him filled up with bromides,” whispered the doctor to
me. “If we can’t get him to sleep he’ll have the D. T.’s.”
Chanler slowly turned his head toward me and endeavored to
open his eyes wide. The effort was too much for him and his face
became distorted with a drunken smile.
“There he is—li’l Gardy, the foe of rum,” he murmured sleepily.
“Model young man. Gardy, know wha’ I’d like see? Like see you
stewed to zenith. Like see you spiff-iflicated. Oh, wha’ ’n ez’bition
you’d be! Horr’ble, horr’ble!” He shook his head slowly. “Nay, nay!
Don’ catch Gardy spiff-iflicated. Don’ catch Gardy putting things in’s
brain to steal his mouth away, do they, Gard’? Noshirr-rr! Noshir-r!
Let George do ’t, eh, Gardy? Let George—let——”
His head fell forward. With an effort he raised it, but his eyes
were closed.
“Gardy—you—you——”
He collapsed slowly upon the pillow and was sound asleep.
Dr. Olson set his glass down and wiped his forehead.
“That’s good,” he said. “But he’s going to be a very sick man.”
“Of course,” I said. “But now that you have got him asleep we are
going to stop his drinking and get him straightened up.”
The doctor looked at Chanler’s puffed face.
“What’s the use?” he said with a shrug of his thin shoulders.
“Besides, he doesn’t want to do anything of the sort.”
 “What he wants doesn’t matter,” I insisted. “He’s got to be
straightened up. What can you do for him?”
The little man looked at me with a weary smile.
“Why this eagerness, Pitt? If I put Chanler on his feet——”
“Then that’s settled,” I interrupted. “You admit you can put him on
his feet, therefore you’ve got to do it. Your word?”
“My word,” he said solemnly, and went to work.
Miss Baldwin was waiting for me as I came from Chanler’s
stateroom.
“I saw you just as you went in,” she said. “Well?”
“He’s sleeping now,” I replied. “He’ll be all right—or, at least better
—when he wakes. George will straighten up.”
She looked at me in that wonderful quiet way of hers.
“Are you so loyal to all your friends, Mr. Pitt?” she said.
“George will straighten up,” I repeated. “He is in Dr. Olson’s
hands. He will make amends when he is himself again.”
She turned away, a wistful—perhaps bitter—smile faintly touching
her lips.
“Miss Baldwin!” I cried apologetically. “Have I said anything to
hurt you, to give you pain?”
“You?” she said, smiling brightly. “Of course you haven’t. How
could you think that? I—I merely happened to think of how different
George was a few months ago. No, no! Don’t grow sad out of
sympathy, please, Mr. Pitt. I’m not unhappy. Do I look it? I cared for
George. I know it now. Maybe I could have learned to care for him
deeply if he had cared for me truly. But he didn’t, and I’m glad I
found it out.”
“You mustn’t say that, Miss Baldwin. You must give him another
chance when he’s himself again.”
“Loyal Mr. Pitt!” she laughed. “Well, I can scarcely help giving
George another chance, can I? Here on the same yacht with him. Mr.
Pitt, I’ll bet I know what you think of me?”
“And that is?”
“That I’m an awful fool to be here?”
 I smiled.
“I knew it!” she cried.
“You’re wrong!” I protested. “I do not think so at this moment.”
“But you have thought so?”
“I have thought you—well, not quite as cautious——”
“Prevaricator! You’ve thought: ‘What sort of a silly madcap is this
girl!’ I know it. Well, I guess you’re right. It was a foolish thing to
do; it’s foolish to be glad at the prospect of adventure. Other girls
wouldn’t do it. They wouldn’t think of it. They’d think a girl queer
who did. That proves it’s foolish, doesn’t it? It isn’t done. I can’t help
it, though; I’ve needed something like this.”
“It is the day of restlessness among American women,” I said
fatuously.
“Restlessness? Is it? Yes, I suppose it is. But my restlessness
doesn’t take the regular, honest truth road, you know. Lots of my girl
friends have felt they wanted to do something, but they’ve wanted
to go suff’ing, or paint, or write, or teach folk-dances, or something
like that. I didn’t, not any more than I wanted to be considered a
doll in pretty clothes all my life.
“I wanted to break away. Well, I did. Here I am. And, scandalous
as it may sound, I’m enjoying every minute. Now, Mr. Pitt, there’s my
whole confession. I have acted foolishly, and I know it, but really, I
feel as if I had broken loose from something that had held me down.
I feel as if it was the beginning of a new life for me—of my real life.”
“A new life?” I said. “Why, that’s what Captain Brack said last
night.”
She looked away.
“Yes, so he did,” she said slowly.
And I thought she shivered a little.

I am afraid I cursed poor George Chanler in unchristian fashion

during the rest of that run up to Kalmut Fiord. For during those days
Captain Brack wooed Miss Baldwin steadily. At each meal he sat at
her side; his eyes were upon her, his magic words were for her
alone. And even while he spoke to her I saw in his eyes that terrible,
ruthless look I knew so well.
“What does the hidden country of Kalmut Fiord hold?” he
speculated one evening. “Ah, Miss Baldwin, if we knew our interest
would be discounted. It is a primitive spot, surely; a primal piece of
earth. Let us pray that it holds Romance, without which there can be
no beginning of a new life.” Once more he repeated: “Hidden
country! There’s some in all of us, and until we explore it we don’t
live.”
The effect of his efforts was apparent upon Miss Baldwin. She
seemed to dread each meeting with him, yet she sat beneath his
spell in a state of fascination. So I cursed poor Chanler. Had he been
the man Miss Baldwin had hoped she would have had no attention
for Brack.
Near dusk on the third day after changing our course we sighted
land over our bows, a tiny gray smudge on the horizon. Our speed
was cut down to a crawl at once. The captain, after studying the
land through his glasses, ordered our course changed to west by
nor’west, and through the thickening darkness we moved at a foot-
pace, gradually drawing nearer a harboring, fir-lined coast line.
That night, while most of us slept soundly, we slipped into Kalmut
Fiord. The cessation of the yacht’s motion aroused me in the
morning, and half awake I dressed and stumbled out on deck to
learn the cause.
In the darkness I had a jumbled impression that the Wanderer
was lying in a small lake surrounded by a circle of small, craggy
mountains. Then, my senses clearing, I realized that I had stepped
into the midst of events of sinister portent.
 XX

It was still too dark to gather an accurate impression of the

yacht’s surroundings, yet light enough to make out what was going
on directly before me. A number of sailors were dropping two of the
port life-boats into the water. They worked eagerly and cautiously,
like men in haste and with a desire for silence. A block, carelessly
handled, swung with a clang against one of the davits and a
subdued voice cursed the guilty man for his clumsiness.
“Don’t do that again.” Through the darkness and morning fog the
whisper sounded like a threat of murder. “Now over with those sea-
ladders.”
The voice was Brack’s.
“All right here Foxy,” said another low voice as the second boat
was dropped with little noise into the water. “Let ’em come.”
This was a new voice to me. It was not Riordan’s nor Garvin’s, nor
Wilson’s, yet it had in it a note of authority which did not belong to
any of the sailors. I was further puzzled because I seemed to have
heard it somewhere before.
“Bring them up, Garvin. Hurry; we’ve got to be up there before it’s
light.”
Brack was speaking again in a loud whisper. Garvin’s great bulk
slipped past me toward the after deck, his feet shuffling along the
deck to make as little noise as possible. He was breathing swiftly and
heavily as a man breathes under the stress of great excitement.
I now saw that the captain was standing at one of the sea-ladders
and at the other was a man whose figure I did not recognize as
belonging to any of the men on board. It was a spare, wiry figure,
with a poise that belonged to no ordinary sailor. I moved a little
closer. Now I saw that the man carried a rifle in the hollow of his
arm. I looked at Brack; he was armed likewise.
That movement proved my undoing.
 “Who the devil’s that?” demanded the wiry man hoarsely.
Brack leaned forward and looked at me steadily for several
seconds.
“Don’t you sleep soundly, Pitt?” he asked.
“Not very,” I replied.
He continued to look at me steadfastly. Presently he began to
grin.
“That is unfortunate for you,” he said at last.
“Surely not,” said I. “Had I been sleeping soundly this morning I
would have missed the sight of all this mysterious preparation.”
He chuckled ominously.
“Had you been sleeping soundly—” he began and stopped. “All
right, men. Hurry.”
A file of men came slipping up from aft. They moved with their
bodies crouched far over and stepped softly. I heard their excited
breathing as they drew near. And each of them bore in his hands a
rifle.
“Four in this boat; four in the other,” commanded Brack. “Get
down there without any noise.”
Garvin started to tumble over the side with the rest of the men;
but Brack stopped him. They whispered together, and Garvin again
went aft.
The men were all in the boats now and Brack and the new man
stood at the ladders waiting to follow. The new man had his back
toward me. He was speaking to the captain.
“Who the devil is this guy, Foxy?” he whispered. “I thought we
were going to make a clean getaway.”
“Pitt,” said Brack, “step up and meet the gold-finder, the man
whose story you didn’t think a good excuse for coming here.”
I stood where I was, but the man turned and took a step forward
to have a better look at me, and then I knew why his voice had
puzzled me. The man was Madigan, whom I had seen quarreling
with Brack back in Billy Taylor’s saloon in Seattle.
 Perhaps some instinct had warned me to be prepared for a shock,
for I looked Madigan over without betraying the rush of thoughts
with which my mind was seething. In a flash the whole of Brack’s
scheming, from the time he had met Chanler in San Francisco to the
present moment, was made plain. He had influenced Chanler to
purchase the Wanderer and go north; he had engaged Madigan to
hide away on board and play the wrecked miner at the proper
moment; he had brought the Wanderer into the bay at night; and he
was now starting out—for what?
I managed to smile as I glanced significantly at the rifles which
both men carried.
“And are you going gold-digging now, Captain Brack?” said I. “I
thought picks and shovels were the proper utensils for mining.”
“Much easier to let others use them,” said he. “Much more
satisfactory to use this—” he patted his rifle—“after others have
used the picks and shovels. As you soon shall see, Mr. Pitt.”
“I——”
He lifted his right hand as if for a signal. Quicker than any normal
thought of mine, instinct whispered the imminence of danger.
I ducked and crouched low before Brack’s signal was completed,
and a fist grazed the top of my head from behind and a hand—
Garvin’s—caught hold of my left arm. Terror drove me to action.
As instinctively as any attacked animal whirls upon its assailant, I
turned on Garvin, sweeping my arms around wildly. He had expected
no resistance, and one of my fists thudded viciously into the middle
of his throat. He gurgled in strange fashion, throwing his head far
back, and I struck him again, struck with a strength which I had not
dreamed that I possessed. I saw him staggering, and turned to run.
Madigan leaped nimbly to block me. I dodged back, but the
captain was there, so I turned to Madigan. He was on me with a
rush; we clinched, struggled, fell, and got up again. This continued
for some time. Then a great weight seemed to drop on the back of
my head and my knowledge of what was happening ceased
suddenly.
 XXI

My next moment of consciousness consisted of a sensation of

helplessness. I was awake; I heard sounds vaguely; but I could not
see, nor could I move.
“There.” A voice seemed to speak from a far-away darkness. “He’s
coming to; you didn’t kill him after all, cap.”
I felt something strike me heavily in the side.
“Yes. He’s coming to. Prod him again. —— him! He delayed us,
and every minute counts.”
Once more the heavy blow fell on my side. I opened my eyes
wearily. Painfully turning my head I looked toward my side and made
out a heavy boot. Some one had been kicking me. My eyes moved
up the boot; Garvin was its owner. The sight of his gross face
brought back memory and consciousness. There was blood on his
mouth; in the lower lip was a long cut, and I was glad.
Garvin was staring at me with a mingling of curiosity and respect
in his expression.
“Where the —— did you learn that punch in the Adam’s apple?”
he said. “That’s a new one to me. And, say, you’re quick; quickest
man I ever see; and you’re all there for a middle-weight, bo.”
“Who hit me in the back of the head?” I demanded weakly. “That
was a cowardly blow.”
I heard a growl somewhere which I recognized as Brack’s.
“We were in a hurry,” he said, “and you would not give us a
chance to handle you gently. You delayed us. That may be serious.”
I strove to rise and struck my chest against a board. I was
conscious of a rhythmic motion, and a dull, squeaky sound, repeated
without cessation. My senses cleared. I turned my head. I was lying
under a seat in one of the life-boats and the boat was being rushed
onward under the impulse of eagerly pulled oars.
 “What’s this?” I groaned. “What sort of an outrage is this?”
I twisted myself from under the seat and sat up, looking around
for the yacht. There was no sight of it. There was no sight of
anything but water and steep hills, and the second life-boat closely
following us. We were pulling up a narrow, winding bay. Its width
was fairly uniform, probably a hundred yards. Its water was pure
blue. And on both sides, and before and behind us, rose the craggy,
fir-clad hills, approaching the size of mountains, shutting us out from
all the rest of the world.
“Sit down, Mr. Pitt; it is more comfortable.” From the bow Brack
spoke, and I turned upon him.
“What do you mean?” I began, and there I stopped.
For, though Brack spoke in laughing fashion, there was no
laughter about his lips, none in his eyes. His face was set like a
bronze mask, his mouth was scarcely visible, his eyes shone hard
and fiery between slitted lids. Brack had ceased to pretend; the
brute in him was having its way, and he didn’t care who saw it.
“You would better have slept soundly this morning, Mr. Pitt,” he
said. “If your foolish fight delayed us too long—you will soon know
why.”
“I want to know why right now!” I cried, in spite of the terror that
his face inspired. “You’ve assaulted me; you’ve taken me off the
yacht by force. You’ll pay for this when we get back home.”
“Suppose,” said he musingly, “suppose you should never get back
home?”
His tone, not his words, froze me. I could not speak. I looked at
the faces of the men who were rowing furiously, at Garvin. And I
looked at the cold blue water through which we were speeding and
knew it was no more remorseless than the men in that boat.
“Don’t you think now it would have been better for you to have
slept?” said Brack.
“I think,” I retorted hotly as the power of speech came rushing
back to me, “that you had better take me back to the yacht; and I
know that I will see you punished for assault for this.”
 A sound like laughter issued from his throat, but his expression
did not change.
“Assault?” he repeated. “Ha! You forget that you are out of the
land of courts now, Pity. Assault! Ha! Why, Pitt, that will be like a
maiden’s kiss compared to what’s going to happen in the next half
hour. Sit down; you’re in that oar’s way. Put him down, Garvin.”
Garvin obediently kicked me back of the knee-joints and I
dropped with a noisy clatter to the bottom of the boat.
“—— you!” swore Brack in a loud whisper. “If you make another
noise like that I’ll have you dumped overboard. You’ve made us late.
Now just you lay still and nice where you are, Pitt; we’re having no
noise on this excursion.”
I sat silent. I was half dazed from the blow on the head and by
my situation, and for the next few minutes I observed what was
taking place as one who is less than half awake. By this time we had
come to the head of the bay and were entering the mouth of a small
river which rambled crookedly down through a gap in the hills.
“More juice in your strokes, men,” whispered Brack. “It’s a strong
current, and we haven’t much farther to go.”
His words stimulated the men. Their fierce eyes grew fiercer, and
they bent to their oars with all their might. Most of them were
panting from excitement and exertion.
“We’ll land here,” said Brack presently. “No noise, men.”
The boats swung in to the bank indicated and the men tumbled
out, clutching their rifles eagerly.
“Come along, Pitt.”
“No,” I responded. “From what I hear you’re bound for some sort
of a crime.”
“So are you. That’s why I took you along—to make you pay for
sleeping so lightly. Get out.”
Two men sprang into the boat toward me, and I was forced to
obey. With Brack in the lead a single file was formed and I started
up a faintly marked footpath which ran along the stream. I was
placed near the middle of the line; Madigan brought up the rear. I
was the only man in the party who was not armed.
 For the next ten minutes we hurried forward, through brush, over
rocks and fallen logs, and through muddy spring-holes without a
word being spoken. Brack in the lead, seemed to take no notice of
the obstacles that presented themselves, and every man in the line
with the exception of myself seemed imbued by the same fierce
eagerness. I was helpless. The man behind me was continually
treading on my heels, his heavy breath was on my neck, and I, too,
was forced to hurry, driven along, moving as in a cruel nightmare.
Brack stopped suddenly and held up his hand. A sound had
broken the silence ahead of us. It was repeated, a dull, slapping
sound, and Brack whispered an oath.
“They’re up; chopping wood for breakfast. Follow me.”
He struck off into a wooded ravine at right angles to the trail. At a
distance which I estimated to be three city blocks from the river he
led the way by zigzags up a series of hills and presently we were
nearing the crest of a ridge beyond which no further hills were
visible.
“Get down now,” he ordered. “The lake’s in the valley over this
hill. The man who shows himself above the brush or makes a noise’ll
get hurt.”
He began to wriggle himself forward through the stunted trees
until at last he was able to peer over the crest of the ridge, and the
rest followed his example.
A small, blackish lake lay in the marshy valley below. On the shore
opposite to us were two log cabins, several huge piles of dirt, and a
crude derrick. Daylight was streaming into the valley, dispersing the
night fogs, and we made out two men moving about the buildings.
Brack swore much but softly.
“Slade and Harris!” He paused to curse again. “—— ’em! We’re
too late. —— you, Pitt, you’ll pay for this.”
“What the ——!” snarled Madigan as the captain hesitated.
“What’s all this foxy work for, Foxy? They’re two and we’re ten. Why
don’t we go down an’ clean ’em up?”
“Easy—easy, Tad,” said Brack softly. “No noise. Slade and Harris
are too good with the rifle to try any straight rushing. Besides,
there’s a back trail over there, and they might get away. They’ve got
the gold cached some place and we may need ’em alive to learn
where it is. A little hanging up by the thumbs will make ’em tell.
Gad! The fools! They’ve got three dumps; that means three shafts.
The thing’s richer than I thought, and they’ve kept it all right down
there because they swore to stay there till they had a hundred
thousand apiece.”
“Gawd!” whispered Garvin. “Let’s take a chance, cap.”
“Easy, Garvin, easy!” chuckled Brack. “They’re a couple of suckers,
but they can shoot.
“Well,” growled Madigan, “let’s have it—when do we go get ’em?”
Brack studied the scene before him for several minutes before
replying.
“We’ve got to wait until they’re in the shafts,” was his decision.
“This is too big a risk, giving ’em a chance. If we jump ’em now from
this side they’ll put up a stiff fight and at the same time have a
chance of getting away over their back trail. And if they get into the
woods, they won’t leave the gold where we can find it easily. We’ve
got to spoil that back trail for ’em.”
“Yep;” said Garvin, “leave ’em no getaway.”
“Madigan,” said Brack, “You take your men and circle around on
this side of the ridge and go north until you strike their trail running
out of the valley.”
“That’ll take a couple of hours.”
“A little longer, probably. When you’re set, fire three shots and
we’ll start to rush ’em from this side. The rest’ll be easy. Boys, by
ten o’clock we’ll all be rich.”
We fell back from the top of the ridge, and in a ravine well out of
sight Madigan led his four men into the forest. Brack waited until
they were out of sight and then hurried us back to the boats. Pulling
Madigan’s boat behind us we were swiftly rowed down the river into
the bay. Here the empty boat was tied up in a well-hidden nook, and
we went on toward the yacht.
I now had an opportunity to note the distance which we had
traveled. The fiord curved raggedly from the river’s mouth toward
the sea. In spite of the foothills which shut us in I saw that our
course at first took us away from the river and the lake. Then, where
the bay began to widen, we began to curve backward until when, at
last the Wanderer, riding serene and white on her cradle of blue
water, appeared before us, I knew that our course had been such
that the distance overland to the miner’s lake could not be much
more than half of what it was by water. I judged the distance down
the bay from the river-mouth to the Wanderer to be about three
miles.
As we made out the yacht in the distance, the Captain looked at
his watch.
“Back in nice time for breakfast,” he said. “Well, Pitt, how does it
feel to belong to a gang of robbers? Please don’t say you don’t
belong. You do, you know; we’ve elected you. Yes; you’re one of us
now, and we’re going to keep close watch on you until this little job
is over.”
“What is your object?” I asked. “Why did you drag me up there
with you?”
“Because I suspect that you like to talk, Pitt,” said he, as he
suddenly changed the course of the boat. “You were unfortunate
enough to see us leaving ship. Had I permitted you to stay on board
you would have talked. You might have talked in alarming fashion,
and I do not wish Miss Baldwin to be alarmed—until our work here is
done, at least.”
“Then why did you bring me back?” I cried. “For you certainly can
not expect me to keep silent after what I have seen and heard.”
“You can talk all you want to now, Pitt,” he laughed. Then I saw
that the boat was pointing toward the shore. “Talk your head off,
Pitt. Because no matter how loud you talk your voice won’t be
among those heard aboard.”
The boat shot into a tiny indentation of the fiord, from which the
Wanderer could not be seen, and grounded on the gravelly beach.
“Will you get out sensibly, Pitt, or will you have to be knocked
down and dragged out?” said Brack carelessly.
I stepped out.
 “Barry, you stay here with him.”
A vicious-looking seaman of medium height followed me onto the
beach, his rifle under his arm.
“We’ll be back in an hour or so,” continued Brack as the boat
backed away. “Must look after our passenger, you know. And be
nice, Pitt, and you won’t get hurt.”
“Yes, and make it —— nice, too!” growled the man Barry, scowling
at me. “’Cause I don’t half like this job an’ I sort o’ figger the cap’
wouldn’t be sore if he come back and found I’d had to put you out
of business.”
 XXII

I stood with my head up until the boat had whisked Brack out of
sight, then slumped down in despair upon a convenient boulder. I
was horrified and frightened. My thoughts had cleared by now and
the full significance of what I had seen, heard, and undergone came
to me. Brutal robbery, probably murder; such was the sum and
substance of Brack’s plans. The expedition and the Wanderer turned
in the tools of a piracy which would have been unbelievable with any
other man than the captain! And Miss Baldwin back there on the
yacht, ignorant of the morning’s happenings, unsuspecting of Brack’s
true character, and I helpless to warn her or be of any assistance.
Brack would keep up the pretense. He would be the smooth-
talking captain this morning as if nothing untoward had happened,
or was going to happen. He would maintain this pose until he had
accomplished the robbery, until it pleased him to drop it. And after
this morning I knew that he would go to any lengths to fulfil his will.
“Cold?” sneered Barry as I shivered. “Well, don’t worry, sissy,
Cap’ll make it warm enough for you when he gets ready to ’tend to
you.”
I turned to plead with him, and he laughed delightedly at the fear
and wretchedness in my face. For I was afraid. This was no place for
me. It was all too strange, too harsh. I was literally sick at my
stomach; and yet I knew all the time that I was going to try to warn
those unsuspecting miners whom Captain Brack planned to catch in
their mine like rats in a pit. Heaven knows I did not wish to do it! In
my heart I protested against the Fate that had placed such a task to
my lot. I was unfit for it. Somebody else, more used to such things,
should have had the job.
I would have pleaded with Barry, have sought to bribe him, but
the expression on his vicious countenance made me hold my tongue.
What could I do? This sort of thing was new to me; how did one go
about it?
I thought of the two miners delving away in their shafts, of them
suddenly looking up to find Brack grinning down at them. The
unfairness of the thing was revolting. Did men do such things to
their fellows in this day and age?
I glanced at Barry and his rifle and knew that they did. Craft and
brutality, those were the laws governing this situation. And craft and
brutality soon began to enter my thoughts as readily as they might
enter those of Brack, Garvin, or the lout who was guarding me.
At my feet lay several stones the size of a man’s fist. Presently I
feigned sleepiness, nodded, and slipped from the boulder to a seat
on the sand.
“Sleepy, eh?” Barry sneered. “You’re a fine piece o’ cheese.”
“I’m sick,” I muttered. “My head aches.”
“Oh, you poor thing!” He prodded me carelessly with the butt of
his rifle. “For two cents I’d give you a clout that’d take the ache out
of that head for good.”
The minutes went by in silence. Half an hour later, perhaps, I saw
Barry’s vigilance begin to relax.
My right hand dropped languidly at my side and found a round
stone, slightly larger than a baseball. Barry did not see.
More time passed. At last Barry, catching himself nodding,
straightened up and again prodded me with the butt.
“Don’t do that again,” I whined. “Please don’t.”
“‘Please don’t!’” mocked Barry.
In his estimation I was such a weakling that he had no need to be
cautious. The rifle-butt again touched my side. I grasped it suddenly
with my left hand, the fingers fastening themselves around the
trigger-guard, and sprang up, the stone in my right hand. Barry
jerked at the rifle, drawing me close, and I felled him to the ground
with a blow from the stone on the temple.
I had the rifle now, and as he strove to rise I struck him on the
head with the heavy barrel and he lay still. I stood over him, ready
to strike again, but he did not move and with the rifle in my hand I
ran through the green-leaved brush which fringed the fiord and
started to climb the rocky hills that walled it in.
What I had to do I knew must be done in a hurry, before Brack or
Madigan were in a position to keep a watch on the lake, and I ran
on, regardless of the fissures and gaps with which the hill was
pitted. In my haste I paid little attention to my path, and near the
top I plunged suddenly through a tangle of brush and fell into what
proved to be the mouth of a cave-like opening in the rocky portion
of the hill.
The cave was so well hidden by the spring foliage that I had
literally to walk into it before suspecting its existence. I hid the rifle
there, clambered out and went on. If my senses of direction and
distance were right the lake should be straight north and about a
mile and a half away.
Though I ran and walked as rapidly as possible, it was half an
hour before I struck the ridge which shut out the lake from sight of
the bay. Then I knew that in spite of my ignorance of the woods, I
had gone straight to my goal. I went down the farther side at once,
keeping myself hidden in the brush as much as possible in case
Madigan’s crew should be on the lookout, and finding the trail along
the river I went straight up toward the miners’ camp.
A man was waiting for me as I stepped from the alder-brush into
the clearing about the mine. My clumsy traveling had warned of my
approach and he lay behind a pile of dirt before a shaft, a large blue
pistol pointing straight down the trail where I emerged.
“Don’t shoot!” I cried running toward him, with my hands in the
air. “I’m a friend. I’ve come to warn you that a man named Brack
with a crew of cutthroats is on his way to raid your camp.”
The mention of Brack’s name had a pitiful effect upon the man.
He leaped back, his eyes shifty with fright, and made as if to run
back to the cabins. He caught himself, however, and swung his pistol
steadily on the trail behind me.
He was an old man with a patriarchal beard and a gentle face.
When he saw that no one was following me he said—
 “Come with me, stranger; we’ll get Bill.”
He retreated, walking backward, covering me and the trail with
his weapon, while I followed. Arriving at the first shaft, still keeping
his eyes on me, he called—
“Oh, Bill!”
A tall, laughing youth, with a soft, curly beard, came clambering
out of the mine in response to his summons. At the sight of me his
hand flashed to the pistol on his hip.
“Tell it to Bill, stranger,” said the patriarch. “Bill, the Laughing
Devil’s back and this gentleman says he’s layin’ to come an’ clean us
pronto.”
“Brack?” gasped the youth, with a frightened glance down the
trail. “Foxy Brack?”
“Yes,” I said. “He’s here to rob you. He’s sent one of his
lieutenants around the ridge to cut off your back trail. He has ten of
the worst men in Christendom with him.”
“Oh, my God!” groaned the young man. Steadying himself he
said, “Who are you, stranger?”
I told about the Wanderer and its party, and about the morning’s
happenings as swiftly as possible.
“Why did you run the risk of coming here and telling us this?”
asked the youth when I concluded. “And how do we know you’re
telling the truth?”
“Bill!” said the old man reprovingly. “Can’t you see? Stranger, we
take this right neighborly of you. My name’s Slade, and this is my
partner, young Bill Harris. Pitt, you said your name was? Well, Mr.
Pitt, you’re a man. This Brack, now, he’s a devil. Bill and me saved
his life when he come ashore up at Omkutsk, and he spoke us fine
and friendly, and acted like a man, and we took him in with us on
this gold find.
“Then one day he tried to put us both out of business and we
caught him in the act just in time. It’s hard to kill a man when you
got him helpless, stranger, though we should ’a’ done it then. We
give him a boat with grub, and when the wind was blowing offshore
we sent him out to sea. The devil must ’a’ took care of its own, since
he’s still living; and now he’s come back to clean us out. We been
sort of ’fraid of it all the time.”
“How many d’ you say with him?” queried young Harris. “And all
bad men, too, eh? God! There’s only two of us——”
“Bill,” said Slade patiently, “we can’t stay an’ fight him. You know
what he is.”
“They’re circling round us now?” Harris was looking around wildly.
“We’re cut off.”
“How many went around to cut our trail, neighbor?”
“Five.”
“We may be able to handle five of ’em, Bill,” said Slade. “We
wouldn’t have no chance with ten. We mustn’t let ’em head us off.
Brack ’ud use fire to make us tell where the gold is cached. We’ll
start right away and travel light.”
Harris ran into the large cabin. I started to go back the way I had
come.
“Wha-a-at? You ain’t going back to Brack’s boat, are you?
Neighbor, there’ll be only hell where that devil is.”
“And for that reason I must go back there.”
“Why?”
“There is a girl—a young lady—on the yacht.”
Old Slade shook his head.
“That dirty devil! But we can’t stay and fight ten men and Brack.
Well, Mr. Pitt, I reckon we owe you our lives and everything we got,
but I dunno how we’re goin’ to square it with you.”
My eyes fell on the automatic pistol in his hand.
“You’re —— whistlin’!” cried Slade suddenly as he thrust the
weapon into my hands. I put it inside my shirt. “That don’t square
us. Best I can do, though. Now, Mr. Pitt—” he gripped my hand
—“God bless yoh!”
 XXIII

I hurried back down the river-trail until I reached the ridge. Here I
quitted the way I had come and climbed away over the hills toward
the sea. My plan was to step aboard the Wanderer while Brack was
absent, and without being seen by any of his men. Hence, I gave
the cove where I had struck down Barry a wide berth. In fact, I did
not follow the windings of the fiord at all but struck straight across
the rough country toward where I judged the sea to be.
I got lost twice. Once I found myself turning toward the fiord and
once I had circled back toward the lake. It was well into the
afternoon when I found the rough seacoast and following it
southward came to the mouth of the fiord and, from a hilltop looked
down upon the Wanderer at anchor.
I saw now why my first impression of the morning had been that
the yacht was surrounded by mountains. This was nearly so. The
hills, one of which I was lying on, walled the fiord in on both sides,
while across its mouth, shutting it in from the sea and leaving only a
narrow channel on either side, lay a narrow, crescent-shaped island
consisting of a fir-covered hill of equal height to those of the
mainland.
The Hidden Country! It was the inevitable name for the region.
Small wonder that Kalmut Fiord was not on the maps. It lay
behind its crescent-shaped island securely hidden from all the world.
Outside, the dun, gray North Pacific heaved and murmured, a part of
the busy world. Somewhere on its restless water ships were sailing,
men were active in the doings of our day and age. But in the hidden
country behind the island there was no such suggestion.
The fiord lay hill-ringed and calm, a part of the world, and yet not
of it. Its green Spring foliage, delicate, masking gray hills and black
cliffs, its quiet blue water, its virgin beaches, its very air, all were
heavy with the primitive’s eternal calm.