45428

Visit https://ebookgate.
com to download the full version and

explore more ebooks
Visualizing Statistical Models and Concepts 1st

Edition R. W. Farebrother
_____ Click the link below to download _____

https://ebookgate.com/product/visualizing-statistical-
models-and-concepts-1st-edition-r-w-farebrother/
Explore and download more ebooks at ebookgate.com

Here are some recommended products that might interest you.
You can download now and explore!
Visualizing Landscape Architecture Functions Concepts

Strategies 1st Edition Elke Mertens
https://ebookgate.com/product/visualizing-landscape-architecture-
functions-concepts-strategies-1st-edition-elke-mertens/
ebookgate.com
Interpreting and Visualizing Regression Models Using Stata

2nd Edition Michael Mitchell
https://ebookgate.com/product/interpreting-and-visualizing-regression-
models-using-stata-2nd-edition-michael-mitchell/
ebookgate.com
Statistical Models 1st Edition A.C. Davison
https://ebookgate.com/product/statistical-models-1st-edition-a-c-
davison/
ebookgate.com
Visualizing Project Management Models and Frameworks for

Mastering Complex Systems 3rd Edition Kevin Forsberg
https://ebookgate.com/product/visualizing-project-management-models-
and-frameworks-for-mastering-complex-systems-3rd-edition-kevin-
forsberg/
ebookgate.com
Statistical Concepts and Applications in Clinical Medicine
1st Edition John Aitchison
https://ebookgate.com/product/statistical-concepts-and-applications-
in-clinical-medicine-1st-edition-john-aitchison/
ebookgate.com
Statistical Misconceptions 1st Edition Schuyler W. Huck
https://ebookgate.com/product/statistical-misconceptions-1st-edition-
schuyler-w-huck/
ebookgate.com
Applied Linear Statistical Models 5th Edition Michael H.

Kutner
https://ebookgate.com/product/applied-linear-statistical-models-5th-
edition-michael-h-kutner/
ebookgate.com
Models for Probability and Statistical Inference Theory

and Applications 1st Edition James H. Stapleton
https://ebookgate.com/product/models-for-probability-and-statistical-
inference-theory-and-applications-1st-edition-james-h-stapleton-2/
ebookgate.com
Models for Probability and Statistical Inference Theory

and Applications 1st Edition James H. Stapleton
https://ebookgate.com/product/models-for-probability-and-statistical-
inference-theory-and-applications-1st-edition-james-h-stapleton/
ebookgate.com
VISUALIZING STATISTICAL
MODELS AND CONCEPTS
R. W. FAREBROTHER
Faculty of Economics and Social Studies
Victoria University of Manchester
Manchester, England
Illustrated by Michaël Schyns
Marcel Dekker, Inc. New York • Basel

TM
Copyright © 2002 by Marcel Dekker, Inc. All Rights Reserved.

ISBN: 0-8247-0718-4
This book is printed on acid-free paper.
Headquarters
Marcel Dekker, Inc.
270 Madison Avenue, New York, NY 10016
tel: 212-696-9000; fax: 212-685-4540
Eastern Hemisphere Distribution

Marcel Dekker AG
Hutgasse 4, Postfach 812, CH-4001 Basel, Switzerland
tel: 41-61-261-8482; fax: 41-61-261-8896
World Wide Web

http://www.dekker.com
The publisher offers discounts on this book when ordered in bulk quantities. For
more information, write to Special Sales/Professional Marketing at the head-
quarters address above.
Copyright # 2002 by Marcel Dekker, Inc. All Rights Reserved.
Neither this book nor any part may be reproduced or transmitted in any form or
by any means, electronic or mechanical, including photocopying, microfilming,
and recording, or by any information storage and retrieval system, without
permission in writing from the publisher.
Current printing (last digit):

10 9 8 7 6 5 4 3 2 1
PRINTED IN THE UNITED STATES OF AMERICA

STATISTICS: Textbooks and Monographs
D. B. Owen
Founding Editor, 1972–1991
Associate Editors
Statistical Computing/ Multivariate Analysis

Nonparametric Statistics Professor Anant M. Kshirsagar
Professor William R. Schucany University of Michigan
Southern Methodist University
Probability Quality Control/Reliability

Professor Marcel F. Neuts Professor Edward G. Schilling
University of Arizona Rochester Institute of Technology
Editorial Board
Applied Probability
Dr. Paul R. Garvey Statistical Distributions
The MITRE Corporation Professor N. Balakrishnan
McMaster University
Economic Statistics
Professor David E. A. Giles Statistical Process Improvement
University of Victoria Professor G. Geoffrey Vining
Virginia Polytechnic Institute
Experimental Designs
Mr. Thomas B. Barker Stochastic Processes
Rochester Institute of Technology Professor V. Lakshmikantham
Florida Institute of Technology
Multivariate Analysis
Professor Subir Ghosh Survey Sampling
University of California–Riverside Professor Lynne Stokes
Southern Methodist University
Time Series
Sastry G. Pantula
North Carolina State University
1. The Generalized Jackknife Statistic, H. L. Gray and W. R. Schucany
2. Multivariate Analysis, Anant M. Kshirsagar
3. Statistics and Society, Walter T. Federer
4. Multivariate Analysis: A Selected and Abstracted Bibliography, 1957–1972, Kocherlakota
Subrahmaniam and Kathleen Subrahmaniam
5. Design of Experiments: A Realistic Approach, Virgil L. Anderson and Robert A. McLean
6. Statistical and Mathematical Aspects of Pollution Problems, John W. Pratt
7. Introduction to Probability and Statistics (in two parts), Part I: Probability; Part II: Statistics,
Narayan C. Giri
8. Statistical Theory of the Analysis of Experimental Designs, J. Ogawa
9. Statistical Techniques in Simulation (in two parts), Jack P. C. Kleijnen
10. Data Quality Control and Editing, Joseph I. Naus
11. Cost of Living Index Numbers: Practice, Precision, and Theory, Kali S. Banerjee
12. Weighing Designs: For Chemistry, Medicine, Economics, Operations Research, Statistics, Kali S.
Banerjee
13. The Search for Oil: Some Statistical Methods and Techniques, edited by D. B. Owen
14. Sample Size Choice: Charts for Experiments with Linear Models, Robert E. Odeh and Martin Fox
15. Statistical Methods for Engineers and Scientists, Robert M. Bethea, Benjamin S. Duran, and
Thomas L. Boullion
16. Statistical Quality Control Methods, Irving W. Burr
17. On the History of Statistics and Probability, edited by D. B. Owen
18. Econometrics, Peter Schmidt
19. Sufficient Statistics: Selected Contributions, Vasant S. Huzurbazar (edited by Anant M.
Kshirsagar)
20. Handbook of Statistical Distributions, Jagdish K. Patel, C. H. Kapadia, and D. B. Owen
21. Case Studies in Sample Design, A. C. Rosander
22. Pocket Book of Statistical Tables, compiled by R. E. Odeh, D. B. Owen, Z. W. Birnbaum, and L.
Fisher
23. The Information in Contingency Tables, D. V. Gokhale and Solomon Kullback
24. Statistical Analysis of Reliability and Life-Testing Models: Theory and Methods, Lee J. Bain
25. Elementary Statistical Quality Control, Irving W. Burr
26. An Introduction to Probability and Statistics Using BASIC, Richard A. Groeneveld
27. Basic Applied Statistics, B. L. Raktoe and J. J. Hubert
28. A Primer in Probability, Kathleen Subrahmaniam
29. Random Processes: A First Look, R. Syski
30. Regression Methods: A Tool for Data Analysis, Rudolf J. Freund and Paul D. Minton
31. Randomization Tests, Eugene S. Edgington
32. Tables for Normal Tolerance Limits, Sampling Plans and Screening, Robert E. Odeh and D. B.
Owen
33. Statistical Computing, William J. Kennedy, Jr., and James E. Gentle
34. Regression Analysis and Its Application: A Data-Oriented Approach, Richard F. Gunst and
Robert L. Mason
35. Scientific Strategies to Save Your Life, I. D. J. Bross
36. Statistics in the Pharmaceutical Industry, edited by C. Ralph Buncher and Jia-Yeong Tsay
37. Sampling from a Finite Population, J. Hajek
38. Statistical Modeling Techniques, S. S. Shapiro and A. J. Gross
39. Statistical Theory and Inference in Research, T. A. Bancroft and C.-P. Han
40. Handbook of the Normal Distribution, Jagdish K. Patel and Campbell B. Read
41. Recent Advances in Regression Methods, Hrishikesh D. Vinod and Aman Ullah
42. Acceptance Sampling in Quality Control, Edward G. Schilling
43. The Randomized Clinical Trial and Therapeutic Decisions, edited by Niels Tygstrup, John M
Lachin, and Erik Juhl
44. Regression Analysis of Survival Data in Cancer Chemotherapy, Walter H. Carter, Jr., Galen L.
Wampler, and Donald M. Stablein
45. A Course in Linear Models, Anant M. Kshirsagar
46. Clinical Trials: Issues and Approaches, edited by Stanley H. Shapiro and Thomas H. Louis
47. Statistical Analysis of DNA Sequence Data, edited by B. S. Weir
48. Nonlinear Regression Modeling: A Unified Practical Approach, David A. Ratkowsky
49. Attribute Sampling Plans, Tables of Tests and Confidence Limits for Proportions, Robert E. Odeh
and D. B. Owen
50. Experimental Design, Statistical Models, and Genetic Statistics, edited by Klaus Hinkelmann
51. Statistical Methods for Cancer Studies, edited by Richard G. Cornell
52. Practical Statistical Sampling for Auditors, Arthur J. Wilburn
53. Statistical Methods for Cancer Studies, edited by Edward J. Wegman and James G. Smith
54. Self-Organizing Methods in Modeling: GMDH Type Algorithms, edited by Stanley J. Farlow
55. Applied Factorial and Fractional Designs, Robert A. McLean and Virgil L. Anderson
56. Design of Experiments: Ranking and Selection, edited by Thomas J. Santner and Ajit C.
Tamhane
57. Statistical Methods for Engineers and Scientists: Second Edition, Revised and Expanded, Robert
M. Bethea, Benjamin S. Duran, and Thomas L. Boullion
58. Ensemble Modeling: Inference from Small-Scale Properties to Large-Scale Systems, Alan E.
Gelfand and Crayton C. Walker
59. Computer Modeling for Business and Industry, Bruce L. Bowerman and Richard T. O'Connell
60. Bayesian Analysis of Linear Models, Lyle D. Broemeling
61. Methodological Issues for Health Care Surveys, Brenda Cox and Steven Cohen
62. Applied Regression Analysis and Experimental Design, Richard J. Brook and Gregory C. Arnold
63. Statpal: A Statistical Package for Microcomputers—PC-DOS Version for the IBM PC and
Compatibles, Bruce J. Chalmer and David G. Whitmore
64. Statpal: A Statistical Package for Microcomputers—Apple Version for the II, II+, and IIe, David G.
Whitmore and Bruce J. Chalmer
65. Nonparametric Statistical Inference: Second Edition, Revised and Expanded, Jean Dickinson
Gibbons
66. Design and Analysis of Experiments, Roger G. Petersen
67. Statistical Methods for Pharmaceutical Research Planning, Sten W. Bergman and John C. Gittins
68. Goodness-of-Fit Techniques, edited by Ralph B. D'Agostino and Michael A. Stephens
69. Statistical Methods in Discrimination Litigation, edited by D. H. Kaye and Mikel Aickin
70. Truncated and Censored Samples from Normal Populations, Helmut Schneider
71. Robust Inference, M. L. Tiku, W. Y. Tan, and N. Balakrishnan
72. Statistical Image Processing and Graphics, edited by Edward J. Wegman and Douglas J.
DePriest
73. Assignment Methods in Combinatorial Data Analysis, Lawrence J. Hubert
74. Econometrics and Structural Change, Lyle D. Broemeling and Hiroki Tsurumi
75. Multivariate Interpretation of Clinical Laboratory Data, Adelin Albert and Eugene K. Harris
76. Statistical Tools for Simulation Practitioners, Jack P. C. Kleijnen
77. Randomization Tests: Second Edition, Eugene S. Edgington
78. A Folio of Distributions: A Collection of Theoretical Quantile-Quantile Plots, Edward B. Fowlkes
79. Applied Categorical Data Analysis, Daniel H. Freeman, Jr.
80. Seemingly Unrelated Regression Equations Models: Estimation and Inference, Virendra K.
Srivastava and David E. A. Giles
81. Response Surfaces: Designs and Analyses, Andre I. Khuri and John A. Cornell
82. Nonlinear Parameter Estimation: An Integrated System in BASIC, John C. Nash and Mary
Walker-Smith
83. Cancer Modeling, edited by James R. Thompson and Barry W. Brown
84. Mixture Models: Inference and Applications to Clustering, Geoffrey J. McLachlan and Kaye E.
Basford
85. Randomized Response: Theory and Techniques, Arijit Chaudhuri and Rahul Mukerjee
86. Biopharmaceutical Statistics for Drug Development, edited by Karl E. Peace
87. Parts per Million Values for Estimating Quality Levels, Robert E. Odeh and D. B. Owen
88. Lognormal Distributions: Theory and Applications, edited by Edwin L. Crow and Kunio Shimizu
89. Properties of Estimators for the Gamma Distribution, K. O. Bowman and L. R. Shenton
90. Spline Smoothing and Nonparametric Regression, Randall L. Eubank
91. Linear Least Squares Computations, R. W. Farebrother
92. Exploring Statistics, Damaraju Raghavarao
93. Applied Time Series Analysis for Business and Economic Forecasting, Sufi M. Nazem
94. Bayesian Analysis of Time Series and Dynamic Models, edited by James C. Spall
95. The Inverse Gaussian Distribution: Theory, Methodology, and Applications, Raj S. Chhikara and
J. Leroy Folks
96. Parameter Estimation in Reliability and Life Span Models, A. Clifford Cohen and Betty Jones
Whitten
97. Pooled Cross-Sectional and Time Series Data Analysis, Terry E. Dielman
98. Random Processes: A First Look, Second Edition, Revised and Expanded, R. Syski
99. Generalized Poisson Distributions: Properties and Applications, P. C. Consul
100. Nonlinear Lp-Norm Estimation, Rene Gonin and Arthur H. Money
101. Model Discrimination for Nonlinear Regression Models, Dale S. Borowiak
102. Applied Regression Analysis in Econometrics, Howard E. Doran
103. Continued Fractions in Statistical Applications, K. O. Bowman and L. R. Shenton
104. Statistical Methodology in the Pharmaceutical Sciences, Donald A. Berry
105. Experimental Design in Biotechnology, Perry D. Haaland
106. Statistical Issues in Drug Research and Development, edited by Karl E. Peace
107. Handbook of Nonlinear Regression Models, David A. Ratkowsky
108. Robust Regression: Analysis and Applications, edited by Kenneth D. Lawrence and Jeffrey L.
Arthur
109. Statistical Design and Analysis of Industrial Experiments, edited by Subir Ghosh
110. U-Statistics: Theory and Practice, A. J. Lee
111. A Primer in Probability: Second Edition, Revised and Expanded, Kathleen Subrahmaniam
112. Data Quality Control: Theory and Pragmatics, edited by Gunar E. Liepins and V. R. R. Uppuluri
113. Engineering Quality by Design: Interpreting the Taguchi Approach, Thomas B. Barker
114. Survivorship Analysis for Clinical Studies, Eugene K. Harris and Adelin Albert
115. Statistical Analysis of Reliability and Life-Testing Models: Second Edition, Lee J. Bain and Max
Engelhardt
116. Stochastic Models of Carcinogenesis, Wai-Yuan Tan
117. Statistics and Society: Data Collection and Interpretation, Second Edition, Revised and
Expanded, Walter T. Federer
118. Handbook of Sequential Analysis, B. K. Ghosh and P. K. Sen
119. Truncated and Censored Samples: Theory and Applications, A. Clifford Cohen
120. Survey Sampling Principles, E. K. Foreman
121. Applied Engineering Statistics, Robert M. Bethea and R. Russell Rhinehart
122. Sample Size Choice: Charts for Experiments with Linear Models: Second Edition, Robert E.
Odeh and Martin Fox
123. Handbook of the Logistic Distribution, edited by N. Balakrishnan
124. Fundamentals of Biostatistical Inference, Chap T. Le
125. Correspondence Analysis Handbook, J.-P. Benzécri
126. Quadratic Forms in Random Variables: Theory and Applications, A. M. Mathai and Serge B.
Provost
127. Confidence Intervals on Variance Components, Richard K. Burdick and Franklin A. Graybill
128. Biopharmaceutical Sequential Statistical Applications, edited by Karl E. Peace
129. Item Response Theory: Parameter Estimation Techniques, Frank B. Baker
130. Survey Sampling: Theory and Methods, Arijit Chaudhuri and Horst Stenger
131. Nonparametric Statistical Inference: Third Edition, Revised and Expanded, Jean Dickinson
Gibbons and Subhabrata Chakraborti
132. Bivariate Discrete Distribution, Subrahmaniam Kocherlakota and Kathleen Kocherlakota
133. Design and Analysis of Bioavailability and Bioequivalence Studies, Shein-Chung Chow and Jen-
pei Liu
134. Multiple Comparisons, Selection, and Applications in Biometry, edited by Fred M. Hoppe
135. Cross-Over Experiments: Design, Analysis, and Application, David A. Ratkowsky, Marc A. Evans,
and J. Richard Alldredge
136. Introduction to Probability and Statistics: Second Edition, Revised and Expanded, Narayan C.
Giri
137. Applied Analysis of Variance in Behavioral Science, edited by Lynne K. Edwards
138. Drug Safety Assessment in Clinical Trials, edited by Gene S. Gilbert
139. Design of Experiments: A No-Name Approach, Thomas J. Lorenzen and Virgil L. Anderson
140. Statistics in the Pharmaceutical Industry: Second Edition, Revised and Expanded, edited by C.
Ralph Buncher and Jia-Yeong Tsay
141. Advanced Linear Models: Theory and Applications, Song-Gui Wang and Shein-Chung Chow
142. Multistage Selection and Ranking Procedures: Second-Order Asymptotics, Nitis Mukhopadhyay
and Tumulesh K. S. Solanky
143. Statistical Design and Analysis in Pharmaceutical Science: Validation, Process Controls, and
Stability, Shein-Chung Chow and Jen-pei Liu
144. Statistical Methods for Engineers and Scientists: Third Edition, Revised and Expanded, Robert
M. Bethea, Benjamin S. Duran, and Thomas L. Boullion
145. Growth Curves, Anant M. Kshirsagar and William Boyce Smith
146. Statistical Bases of Reference Values in Laboratory Medicine, Eugene K. Harris and James C.
Boyd
147. Randomization Tests: Third Edition, Revised and Expanded, Eugene S. Edgington
148. Practical Sampling Techniques: Second Edition, Revised and Expanded, Ranjan K. Som
149. Multivariate Statistical Analysis, Narayan C. Giri
150. Handbook of the Normal Distribution: Second Edition, Revised and Expanded, Jagdish K. Patel
and Campbell B. Read
151. Bayesian Biostatistics, edited by Donald A. Berry and Dalene K. Stangl
152. Response Surfaces: Designs and Analyses, Second Edition, Revised and Expanded, André I.
Khuri and John A. Cornell
153. Statistics of Quality, edited by Subir Ghosh, William R. Schucany, and William B. Smith
154. Linear and Nonlinear Models for the Analysis of Repeated Measurements, Edward F. Vonesh
and Vernon M. Chinchilli
155. Handbook of Applied Economic Statistics, Aman Ullah and David E. A. Giles
156. Improving Efficiency by Shrinkage: The James-Stein and Ridge Regression Estimators, Marvin
H. J. Gruber
157. Nonparametric Regression and Spline Smoothing: Second Edition, Randall L. Eubank
158. Asymptotics, Nonparametrics, and Time Series, edited by Subir Ghosh
159. Multivariate Analysis, Design of Experiments, and Survey Sampling, edited by Subir Ghosh
160. Statistical Process Monitoring and Control, edited by Sung H. Park and G. Geoffrey Vining
161. Statistics for the 21st Century: Methodologies for Applications of the Future, edited by C. R.
Rao and Gábor J. Székely
162. Probability and Statistical Inference, Nitis Mukhopadhyay
163. Handbook of Stochastic Analysis and Applications, edited by D. Kannan and V. Lak-
shmikantham
164. Testing for Normality, Henry C. Thode, Jr.
Additional Volumes in Preparation
Handbook of Applied Econometrics and Statistical Inference, edited by Aman Ullah, Alan T. K.
Wan, and Anoop Chaturvedi
Visualizing Statistical Models and Concepts, R. W. Farebrother and Michael Schyns

Preface
Until relatively recent times, geometrical and mechanical analogies

played an essential role in the development of mathematical
models for the physical sciences. The same is true of the models
and concepts of mathematical statistics. However, from the begin-
ning of the twentieth century, there has been a growing tendency
for statisticians to turn away from such matters, so much so that
typical modern statisticians have few, if any, geometrical insights
into the nature of the fitting problems they are dealing with, and no
mechanical insights whatsoever.
This almost total absence of geometrical and mechanical ana-
logies from modern mathematical statistics constitutes a sad deple-
tion of the subject. For, although visual representations of this
type are no longer required as the basis of suitable analogue com-
puting devices, they are still valuable as a means of enhancing the
practitioner’s understanding of the concepts involved. In this
book, I attempt to remedy this parlous situation by suggesting
suitable geometrical and mechanical analogies for some of the
more accessible statistical techniques. In particular, I address the
problem of determining the position of a multivariate location
parameter and the problem of fitting a plane or curved surface
to multivariate data when the goodness of fit criterion is variously
defined by the sum of the squared deviations, the sum of the abso-
iii
iv Preface
lute deviations, the median absolute deviation, or the maximum

absolute deviation.
This book is intended for use as auxiliary reading by students
of mathematical statistics and as a reference source for research
workers in this area. It may also serve as the basis of a (largely
heuristic) introductory course in statistics for students of engineer-
ing, physics, and technology who are sufficiently familiar with the
mechanical concepts underlying this treatment of the subject.
To enhance its role as a reference work, the successive chap-
ters of the book have been made as independent of one another as
possible. However, I have reluctantly decided not to investigate
any of the linear or nonlinear programming procedures in detail,
as to do so would presume too much of readers’ knowledge of this
area and would overburden the text with numerical examples.
Hopefully, figures similar to those presented in the text will
soon make their appearance in commercially produced packages of
statistical graphics.
The successive drafts of this book were typed by my wife,
Sheila, and myself. The present version of the book was prepared
from our final typescript by Keyword Typesetting Services under
the direction of Marcel Dekker, Inc. I am indebted to Stephen
Pollock of Queen Mary and Westfield College, University of
London, England, for translating the original draft into Plain
TEX; to Götz Trenkler and Sven-Oliver Troschke of the
University of Dortmund, Germany, for their detailed criticism of
a later version; but, above all, I am indebted to Michaël Schyns of
the University of Namur (FUNDP), Belgium, for creating the
figures given in the text, and to my wife, Sheila, for all her help
in the preparation of this book, not least for her discovery of a
person skilled in statistical graphics amongst those attending the
Third L1 -norm Conference in Neuchâtel, Switzerland, in August
1997. Without their aid this project would surely have foundered.
R. W. Farebrother
Contents
Preface iii
CHAPTER 1
Introduction 1
1.1 Introductory Remarks 1

1.2 The Role of Geometrical Models in Statistics 3
1.3 The Analogy Implicit in Some Statistical
Nomenclature 5
1.4 A Simple Mechanical Model for the Arithmetic
Mean 8
1.5 Mechanical Models in Science 10
1.6 Possible Uses of this Book 10
References 11
v
vi Contents
CHAPTER 2
Abstract Geometrical and Mechanical Representations 13
2.1 Bayesian and non-Bayesian Likelihood Functions 13

2.1.1 Likelihood Functions as Abstract Potential Energy
Functions 13
2.1.2 Hypothesis Tests Based on Differences in Energy
Levels 15
2.2 Geometrical Representations in Prediction Space 16
2.2.1 Prediction Space Representation 17
2.2.2 A Geometrical Analogy for the Correlation
Coefficient 18
2.2.3 Planes and Hyperplanes in Prediction Space 19
2.3 Observation Space Representation 21
2.4 Parameter Space Representation 22
2.5 Line Fitting by Eye 25
References 28
CHAPTER 3
Mechanical Models for Multidimensional Medians 29
3.1 Mechanical Models for One-Dimensional Medians 29

3.1.1 Potential Energy Models for One-Dimensional
Medians 29
3.1.2 Balance of Forces in One-Dimensional Medians 33
3.2 Mechanical Models for Two-Dimensional Medians 35
3.2.1 Potential Energy Models for Two-Dimensional
Medians 35
3.2.2 Geometrical Solution in Parameter Space 36
3.2.3 Vectorial Representation of Forces 37
3.2.4 Particular Solutions 42
3.2.5 Algorithm for the Mediancentre 43
3.2.6 Non-unit Weights 45
3.3 Linear, Curvilinear, and Regional Constraints 46
3.3.1 Linear and Curvilinear Constraints 46
Contents vii
3.3.2 Regional Constraints 49

3.3.3 Hypothesis Tests 50
3.4 Three Further Generalisations of the Mechanical
Model 52
3.4.1 Models with Non-Euclidean Distances 52
3.4.2 Two or More Medians 52
3.4.3 Transportation Network with Multiple Nodes 53
3.5 Mechanical Models in Sums of Areas 55
3.5.1 Soap Bubble Models 55
3.5.2 Oja’s Spatial Median 58
3.5.3 Functional Approximation 59
References 60
CHAPTER 4
Method of Least Squared Deviations 63
4.1 One- and Two-Dimensional Means 63

4.1.1 Mechanical Models for One-Dimensional Means 63
4.1.2 Balance of Forces in the One-Dimensional Case 65
4.1.3 Mechanical Model in the Two-Dimensional Case 66
4.1.4 Balance of Forces in the Two-Dimensional Case 67
4.1.5 The Modulus of the Parameter Estimates 68
4.2 Linear, Curvilinear, and Regional Constraints 70
4.3 Simple Linear Regression 76
4.3.1 Orientation 76
4.3.2 Mechanical Model and Balance of Forces in the
Conventional Case 78
4.3.3 Mechanical Model in the Orthogonal Case 80
4.3.4 Balance of Forces in the Orthogonal Case 82
4.4 Moduli of Parameter Estimates 83
4.4.1 Direct Evaluation 83
4.4.2 Oblique Transformation 84
4.4.3 Relative Equilibrium 86
4.5 Influential Observations 86
4.5.1 Changing Moduli 86
4.5.2 Adding Observations 87
viii Contents
4.5.3 Measures of Influence 87

4.5.4 Ridge Regression 88
4.6 Linear Restrictions and Hypothesis Tests 89
4.6.1 Linear Constraints 89
4.6.2 Hypothesis Test 90
4.6.3 Slope Test 92
4.7 Inertial Models 93
4.8 Matrix Representation 94
References 94
CHAPTER 5
Method of Least Absolute Deviation 97
5.1 One- and Two-Dimensional Medians 97

5.1.1 Mechanical Models for One-Dimensional Medians 97
5.1.2 Mechanical Models for Higher-Dimensional
Medians 100
5.2.1 Conventional Line Fitting Problem 101
5.2.2 Orthogonal Line Fitting Problem 106
5.2.3 Linear Constraints 107
5.2.4 Hypothesis Tests 108
5.3 Parameter Space Representation and Duality 110
5.3.1 Simple Mechanical Model in Parameter Space 110
5.3.2 Perpendicular Models in Parameter Space 111
5.3.3 Mechanical Models for Directional Data 114
5.4 Geometrical Solution 114
5.4.1 Geometrical Plot of the Optimality Function 114
5.4.2 Edgeworth’s Double Median Method 115
5.4.3 Edgeworth’s Iterative Method 120
5.4.4 An Improved Iterative Method 120
5.4.5 Linear Programming Formulation 121
5.4.6 Applications to Voting in Committees 122
5.5 Elemental Set Characterisation of Solutions to Fitting
Problems 125
5.5.1 Elemental Set Solutions 125
Contents ix
5.5.2 Elemental Set Characterisation of the Weighted Least

Squares Solution 127
5.5.3 Explicit Characterisation of the Weighted Least
Squares Solution in the Two-Dimensional Case 128
5.5.4 The Lp -norm Criterion 130
5.5.5 Elemental Set Approximations 131
5.6 Prediction Space Representation 132
5.6.1 Prediction Space Representation 132
5.6.2 The One-Dimensional Fitting Problem 136
5.6.3 Higher Dimensional Models 139
References 140
CHAPTER 6
Minimax Absolute Deviation Method 143

6.1.1 Observation Space Representation 143
6.1.2 Mechanical Model Based on Strings and Blocks 145
6.1.3 Parameter Space Representation 146
6.2.1 Geometrical Solution in Observation Space 147
6.2.2 Influential Observations 150
6.2.3 Geometrical Representation in One-Dimensional
Parameter Space 151
6.2.4 Geometrical Representation in Two-Dimensional
Parameter Space 152
6.2.5 Simplified Geometrical Solution in Two-Dimensional
Parameter Space 156
6.2.6 Linear Programming Formulation 158
6.3 Geometrical Representation of the Harmonic
Model 159
References 160
x Contents
CHAPTER 7
Method of Least Median of Squared Deviations 163

7.1.1 Heuristic Derivation of Least Median of Squared
Deviations Criterion 163
7.1.2 One- and Two-Dimensional Means 164
7.1.3 Parameter Space Representation 167
7.2.1 Physical Solution in Observation Space 169
7.2.2 Influential Observations 170
7.2.3 Geometrical Representation in Parameter Space 170
7.2.4 An Alternative Geometrical Solution 172
7.2.5 A Second Geometrical Solution 172
7.2.6 Nonlinear Programming 174
7.3 Generalisation to Minimum Volume Median Ellipsoids
and Ellipsoidal Cylinders 175
References 175
CHAPTER 8
Mechanical Models for Metric Graphs 177
8.1 Introduction 177

8.2 Metric Graphs 178
8.3 Pairwise Preference Orderings 181
8.4 Transitive Preference Orderings 183
8.5 Three-Dimensional Models 186
8.6 Four-Dimensional Models 188
8.7 Further Generalisations 189
References 190
Contents xi
CHAPTER 9
Categorical Data Analysis 191
9.1 Nature of Categorical Data 191

9.2 Hydrostatic Models 194
9.2.1 Static Model 194
9.2.2 Balance of Forces 196
9.2.3 Potential Energy Analysis 197
9.2.4 Hypothesis Test of Independence 198
9.3 Gas-Pressure Models 199
9.3.1 Mechanical Model 199
9.3.2 Potential Energy Analysis 201
9.3.3 Hypothesis Tests of Independence 202
9.3.4 Analogy with the Chemical Balance Model 203
9.4 Two-Way Tables 204
9.5 Hydrostatic Models in Economics 206
References 206
CHAPTER 10
Method of Averages and Curve Fitting by Splines 207
10.1 Mechanical Models for Multivariate Means 207

10.1.1 One-dimensional Means 207
10.1.2 Multidimensional Means 208
10.2 The Method of Averages 209
10.3 Smoothing by Linear and Cubic Splines 212
10.3.1 Linear Splines 212
10.3.2 Cubic Splines 215
10.4 Multidimensional Medians 217
References 218
xii Contents
CHAPTER 11
Multivariate Generalisations of the Method of Least

Squares 219
11.1 Multivariate Statistical Analysis 219

11.2 Orthogonal Least Squares and Principal
Components 219
11.2.1 Orthogonal Least Squares 219
11.2.2 Fitting Elliptical Contours 223
11.2.3 Fitting a Circle 224
11.3 Procrustes Rotation 226
11.3.1 Orthogonal Procrustes Rotation 226
11.3.2 Fitting Spherical Data 227
11.4 Multidimensional Scaling 229
11.5 Concluding Remarks 231
References 231
List of Figures 233
List of Tables 241
Name Index 243
Subject Index 245

CHAPTER 1
Introduction
1.1 Introductory Remarks

Geometrical diagrams are a familiar feature of elementary statis-
tical textbooks. Amongst the most familiar of such diagrams are
the standard (side-by-side) and the partitioned (end-to-end) bar
charts or histograms of Figures 1.1 and 1.2 and the (observation-
space) representations of the conventional and orthogonal linear
regression models of Figures 1.3 and 1.4.
But these familiar representations of statistical procedures are
entirely static and can only benefit from the admixture of a certain
amount of mechanical interpretation. However, such interpreta-
tions need to be chosen with care if the resulting combination is
to offer the clearest possible insight into the essential nature of the
relevant statistical procedure.
The psychological importance of clear visual impressions of
the mathematical concepts underlying particular theories has been
stressed by Friendly (1995), Sall (1991a, 1991b), and Skemp (1971).
In this book, we shall find that many of the important concepts of
mathematical statistics can be associated with physical models; and
that the optimality criteria of statistical estimation procedures can
often be interpreted in terms of the concept of potential energy.
For example, in Chapters 4 and 5 we shall find that our intuitive
understanding of the linear regression model of Figures 1.3 and 1.4
is significantly enhanced if we interpret the line segments joining
1
2 Chapter 1
Figure 1.1 Side-by-side bar chart.
the observed points to the fitted line as if they were springs or

strings under tension connected to a rigid rod. Similarly, in
Chapter 9, we shall find that our understanding of the dynamic
model underlying the bar charts or histograms of Figures 1.1 and
1.2 is much improved if we interpret the rectangular areas in these
figures as if they were the cross-sections of adjacent containers
holding appropriate quantities of liquid or gas.
In this context, we should also mention that there is a class of
complex mathematical problems that are difficult to solve without
the insight gained from appropriate mechanical models. The prob-
lems discussed in this book are not of this type. Nevertheless, at the
end of Chapter 3, we shall take the opportunity of illustrating this
point by giving a brief outline of Richard Courant’s (1940) use of
Figure 1.2 End-on-end bar chart.

Introduction 3
physical analogies in solving a class of difficult problems in the

calculus of variations.
1.2 The Role of Geometrical Models

in Statistics
Examining the simple linear regression problem in greater detail,
we find that we are given a set of n observations on two statistical
variables X and Y. From the perspective of analytical geometry, it
is natural to represent these n pairs of observations by a set of n
points in a two-dimensional Cartesian plane defined by a pair of
orthogonal axes marked with appropriate ranges of values of X
and Y. Similarly, the hypothetical linear relationship underlying
the observations on these two variables can be represented by a
straight line in the same Cartesian plane. For any straight line in
this plane, we may measure the distances from the points to the
line, and we may define a goodness of fit criterion which is an
increasing function of the absolute values of these distances.
Our problem is to identify a line that minimises the value of
the selected optimality criterion. In particular, if the optimality
criterion is the sum of the squared deviations of the points from
the line, and if these deviations are measured perpendicular to the
x-axis, then the best fitting line is the usual least squares line, see
Figure 1.3.
Figure 1.3 Conventional line fitting problem.

4 Chapter 1
On the other hand, if distances are measured perpendicular to

the fitted line, then the best fitting line is the orthogonal least
squares line, see Figure 1.4.
A second application of geometry to the regression problem
arises when the elliptical contours of the bivariate normal distribu-
tion of X and Y is directly observed, or we have sufficiently many
observations to reconstruct these elliptical contours. In this con-
text, the orthogonal regression line described above corresponds to
the major axis of the elliptical contours, see Figure 1.5.
To generate the usual y on x regression line, we begin by
finding points of tangency between the ellipses and lines drawn
parallel to the y-axis. We note that these points of tangency lie
on a straight line which is known as the y on x regression line, see
Figure 1.6.
The alternative x on y regression line can be generated in a
similar way by considering the points of tangency to a set of lines
drawn parallel to the x-axis, see Figure 1.7.
In elementary statistical textbooks, the directions indicated by
the x- and y-axes are often referred to as the horizontal and vertical
directions respectively. However, it is not convenient to use this
terminology in the present book as we shall be concerned with
models that involve the representation of the optimality function
as a three-dimensional geometrical figure lying vertically above the
(horizontal) plane defined by these two axes. For this reason, the
vertical direction will be strictly reserved for the representation of
deviations from the fitted relationship and closely related func-
Figure 1.4 Orthogonal line fitting problem.

Introduction 5
Figure 1.5 Orthogonal regression problem.
tions. The y-axis will usually be said to run in a south–north direc-

tion in the horizontal plane and the x-axis in a west–east direction.
Thus, the usual nomenclature will only be employed when a func-
tion of the observed deviations is plotted against a single explana-
tory variable, as occurs in the opening subsections of Chapters 3
and 4.
1.3 The Analogy Implicit in Some

Statistical Nomenclature
Having briefly examined the role of several geometrical concepts in
statistical analyses, we can proceed to outline four statistical con-
cepts that originate in geometry and mechanics.
Figure 1.6 Conventional y on x regression problem.

6 Chapter 1
Figure 1.7 Conventional x on y regression problem.
Density: The concept of the density of a statistical distribu-

tion is readily interpreted in geometrical terms. Suppose that we
wish to compare the density of the populations in two or more
towns. We may begin by constructing a set of plane figures whose
areas are proportional to the land areas of the corresponding
towns. These plane figures may be scale representations of the
geographical shape of the towns or they may be stylised represen-
tations in the form of discs or squares. The shape of the plane
figures is immaterial. What is essential is that the area of the figures
should be proportional to the areas of the associated towns.
Inside each figure we place a set of points whose numbers are
proportional to the size of the population of the corresponding
towns. The points should be of equal size and as evenly spaced
as possible. The density of the points will show the densities of the
populations of the corresponding towns. The more densely packed
the points are in the available areas, the more densely packed will
the corresponding populations be in reality.
Thus, a visual inspection of Figure 1.8 immediately reveals
that the town represented by the rectangle on the right covers a
greater area than does the town represented by the rectangle on the
left. Further, the population of the town on the right is more
crowded together than that of the town on the left, as the points
are more densely packed into the rectangle on the right. In addi-
tion, for the particular instance illustrated here, it is possible to
deduce that the population of the town on the right is greater than
that on the left. On the other hand, if the town represented by the
Introduction 7
Figure 1.8 Relative densities of two populations.
smaller rectangle had had the higher density, then it would have
been necessary to count the points in the rectangles before the
populations of the two towns could be compared.
Weight: The idea of the weight of an observation is quite
old. It features in the early work on the method of least squares
of Carl Friedrich Gauss and Pierre Simon Laplace dating from
the beginning of the nineteenth century. The idea was
expounded in greater detail by George Boole in his account of
the language of chance in 1857. [We shall not give specific refer-
ences to these and later historical remarks. Interested readers are
referred to Farebrother (1999), Hald (1998), and Stigler (1986)
for details.]
As we might expect, an observation with greater weight has
more influence on the estimated values of the parameters of a
model than does one of lesser weight. The term weight also denotes
the level of certainty, and the more weight attached to the estimate
8 Chapter 1
of a parameter value, the less will this value be affected by sub-

sequent observations.
Laplace also employed the concept of weight in his analysis of
the problem of minimising the sum of the absolute deviations. We
shall tend to follow mechanical usage and reserve the word weight
for the problem of least absolute deviations, see Chapter 5. We
shall use the alternative word modulus in the context of the least
squares problem of Chapter 4. However, we shall not illustrate this
concept here. Instead, we refer readers to the next section and to
Chapters 3 and 5 for the relevant diagrams.
Moment: If we are given a set of observations, represented as
points distributed along a line, together with an arbitrary point on
the same line, then we may rotate the given points about an axis
passing through the arbitrary point. The more disperse is the dis-
tribution of the points, the more reluctant will the system be to
rotate about the chosen axis under the influence of a given rota-
tional force. The degree of inertia of the system is quantified by the
sum of the squares of the distances of the points from the axis of
rotation. This sum is known as the second moment or moment of
inertia of the distribution. Moments of other orders are defined in
a similar way by summing the relevant power of the distances from
the given points to the axis.
Degrees of Freedom: The number of degrees of freedom of a
particular problem records the number of entities that may be
varied independently of each other. Like the concepts of weight
and moment, this concept has been adopted from physics.
1.4 A Simple Mechanical Model for

the Arithmetic Mean
As a representation of the arithmetic mean, we consider a chemical
balance in a state of equilibrium. Suppose that we are given a set of
n observations on a single variable Y. Then we may represent these
observations as a scatter of points on the horizontal y-axis, and we
may treat this axis as a horizontal beam with unit weights attached
at the given points. Our problem is to place a fulcrum at the point
of balance of the weighted beam.
Introduction 9
Wherever we place the fulcrum, the weights to the right of that

point will cause the beam to rotate in a clockwise direction, whilst
weights to the left will cause it to rotate in an anticlockwise direc-
tion. The turning effect of a weight is measured by the product of
the weight with its horizontal distance from the fulcrum, and is
known as a couple.
Given the position of the fulcrum, we may determine the net
clockwise couple of the system as a whole by summing the couples
associated with each of the n weights. If this sum is positive then
the beam will rotate in a clockwise direction, and if negative then it
will rotate in an anticlockwise direction. Clearly, the system will be
in equilibrium when the fulcrum is placed at the point which sets
this sum equal to zero. The point of balance of the weighted beam
corresponds to the arithmetic mean of the observations.
Figures 1.9a and 1.9b illustrate the case of a beam loaded with
two unit weights. In the situation illustrated in Figure 1.9a, the
fulcrum has been placed near the weight on the left and the beam
will tend to rotate in a clockwise direction about this point. By
contrast, in the situation illustrated in Figure 1.9b, the fulcrum has
been placed exactly halfway between the two weights and the beam
will remain in its present position.
This mechanical model may be generalised to yield valuable
representations of weighted or unweighted multivariate means, see
Chapter 10. However, it is unsatisfactory as a general analogy
since it cannot be applied to more advanced statistical techniques
without establishing a dynamic concept of which the net clockwise
couple is the derivative. Even then, it is not easy to accommodate
fitting procedures that are more general than the method of ortho-
gonal least squares. See Chapter 4 for our preferred quasi-static
model of the arithmetic mean.
Figure 1.9 A loaded beam (a) when not in equilibrium, and (b) when in
equilibrium.
10 Chapter 1
1.5 Mechanical Models in Science

Like the chemical balance of Section 1.4, almost all of the physical
models discussed in later chapters are drawn from classical
mechanics. However, we are not restricted to analogies from this
source as several of the physical sciences exhibit similar features.
Indeed, Franksen (1969) has identified a group of quasi-static
models from classical mechanics, electrical network theory, and
thermodynamics (which he renames thermostatics), whose basic
concepts may be employed as analogies for the corresponding con-
cepts of other sciences in this group, and also for the correspond-
ing models of economics and statistics.
Although, in this book, we shall concentrate on analogies
drawn from classical mechanics, we shall mention models drawn
from hydrostatics in Chapter 9. In principle, we may extend our
scope to include models based on the flow of liquids in systems of
pipes or on the flow of currents in electrical networks. But, such
preliminary analyses as we have been able to conduct with hydro-
dynamic and electrical network theory suggest that these subject
areas cannot offer significant insights into the workings of
statistical models. Models of this type are, however, employed
in economics where they are associated with the names of
François Quesnay and Alban William Phillips. Similarly, we may
base our models on systems of particles in motion under the effect
of mutual gravitation; or, indeed, on collections of molecules in
equilibrium as crystalline structures under a balance of forces. But,
any attempt to address these models would take the analysis
beyond the competence of the author, if not that of the reader.
Thus, although it is possible to base statistical models on a
wide class of paradigms suggested by a variety of physical pro-
cesses, we shall concentrate on those approaches that seem most
accessible and potentially most valuable to students of statistics.
1.6 Possible Uses of This Book

At first sight, it might seem that this book is specifically addressed
to students and research workers who have followed a conven-
Introduction 11
tional course in mathematical statistics with little or no support

from geometrical or mechanical models. And, viewed in this light,
it will serve as the basis of a supplementary course at the graduate
or advanced undergraduate level for conventionally trained statis-
ticians.
However, a more detailed examination of the contents will
establish that there are two sub-themes within the unifying frame-
work offered by the concept of potential energy, namely those
associated with the use or non-use of the differential calculus
and projective geometry duality. If the latter aspect of the subject
is eliminated, then the remaining material in Section 2.1, and in
Chapters 1, 3, 4, 9, 10, and 11 could be used as the basis of an
elementary introduction to statistical estimation procedures for
engineers, physicists, technologists, and others with a solid ground-
ing in classical mechanics.
Alternatively, if we omit any mention of the differential cal-
culus, then the remaining material in Chapters 1, 2, 3, 5, 6, 7, and 8
can be presented in the form of a geomechanical supplement to a
conventional discussion of the line fitting problem for students of
linear and nonlinear programming.
In addition, some of the mechanical models described in this
book may be suitable for use by professional statisticians in their
consultancy work when trying to explain some of their more
abstruse estimation problems to clients with little knowledge of
statistics. For instance, the contrast between the results generated
by the spring models of Chapter 4 and the string models of
Chapter 5 may help to justify the use of a more robust fitting
procedure than the method of least squares. Similarly, the size of
the strain in the springs of the models of Chapter 4 or the tension
in the strings of the models of Chapter 5 may help to demonstrate
the unreasonableness of some of the heuristic constraints cham-
pioned by some clients.
References
Courant, R. (1940), Soap film experiments with minimal surfaces,
American Mathematical Monthly 47: 167–174.
12 Chapter 1
Farebrother, R. W. (1987), Mechanical representations of the L1

and L2 estimation problems, in Y. Dodge (Ed.), Statistical
Data Analysis Based on the L1 -Norm and Related Methods,
North-Holland Publishing Company, Amsterdam, 455–464.
Farebrother, R. W. (1999), Fitting Linear Relationships: A History
of the Calculus of Observations 1750–1900, Springer-Verlag,
New York.
Franksen, O. I. (1969), Mathematical programming in economics
by physical analogies, Simulation June, 297–314; July, 25–42;
August, 63–87.
Friendly, M. (1995), Conceptual and visual models for categorical
data, The American Statistician 49: 153–160.
Hald, A. (1998), A History of Mathematical Statistics from 1750 to
1930, John Wiley and Sons, New York.
Sall, J. (1991a), The conceptual model behind the picture, ASA
Statistical Computing and Statistical Graphics Newsletter 2:
5–8.
Sall, J. (1991b), The conceptual model for categorical responses,
ASA Statistical Computing and Statistical Graphics Newsletter
3: 33–36.
Skemp, R. R. (1971), The Psychology of Learning Mathematics,
Penguin Books, Harmondsworth, England.
Stigler, S. M. (1986), The History of Statistics: The Measurement of
Uncertainty Before 1900, Harvard University Press,
Cambridge, Massachusetts.
CHAPTER 2
Abstract Geometrical and
Mechanical Representations
2.1 Bayesian and non-Bayesian

Likelihood Functions
2.1.1 Likelihood Functions as Abstract
Potential Energy Functions
In this chapter, we initiate our discussion of a wide range of dis-
parate statistical, geometrical, and mechanical concepts which will
prove useful to a greater or lesser degree in subsequent chapters.
The first of these concepts is the (Bayesian or non-Bayesian) like-
lihood function and the associated hypothesis test.
Suppose that X is a random variable with a continuous prob-
ability density function f ðx; Þ characterised by the value of a single
unknown parameter . Then the probability density associated
with a particular value X ¼ xi of this variable is given by the
corresponding value of the function
f ðxi; Þ
and the joint probability density of a set of n observations x1 ; x2 ;

. . . ; xn by the likelihood function
Y
Lð; x1 ; x2 ; . . . ; xn Þ ¼ f ðxi ; Þ
i
13
14 Chapter 2
In certain circumstances, this function of x1 ; x2 ; . . . ; xn given

the value of is converted into a function of given the observed
values of x1 ; x2 ; . . . ; xn by (implicitly or explicitly) multiplying it by
a function embodying the prior information on . The posterior
density function obtained in this way is employed when one wishes
to conduct a Bayesian analysis of the model.
For a non-Bayesian analysis of the same model, the function
L itself is treated as if it were a function of the parameter given
the observations x1 ; x2 ; . . . ; xn (rather than as a function of the
observations given the parameter). In this context, the unknown
parameters are estimated by choosing values for them in such a
way as to maximise the likelihood function.
For example, in the special case of the normal distribution
with mean and variance unity, we find that X has density func-
tion
1 1 2
f ðxi ; Þ ¼ pffiffiffiffiffiffi e 2 ðxi Þ
2p
and that has a likelihood function
Y 1 1 P 2
f ðxi ; Þ ¼ pffiffiffiffiffiffi e 2 ðxi Þ
2p
This likelihood function is clearly maximised when the scaled sum
of squared deviations
1X
ðxi Þ2
2
is minimised.
William Fishburn Donkin (1844) gave a mechanical interpre-
tation of this procedure by implicitly suggesting that we may
regard the negative of the logarithm of the likelihood function as
if it were a potential energy function. Moreover, this interpretation
of the sum of the squared deviations function as a potential energy
function may be extended to a wide range of traditional optimality
criteria which need not be related to probability density functions.
In particular, the sum of the pth powers of the absolute deviations
criterion
X
jxi jp
may be interpreted as a potential energy function for all positive

values of the parameter p. Further, this optimality criterion may be
derived from the density function of a symmetric stable law if (and
only if) the value of p lies in the range 0 < p 2.
An immediate consequence of this interpretation of the
chosen optimality criterion as a potential energy function is that
we may interpret the derivative of this function with respect to the
parameter values as the negative of a force function acting on the
current parameter values. This aspect of the subject will be
developed further in later chapters.
2.1.2 Hypothesis Tests Based on

Differences in Energy Levels
In Subsection 2.1.1 we have seen that the sum of squared devia-
tions optimality function may be interpreted as a potential energy
function. In general, the optimal values of the parameters are
determined by the unconstrained minimum of this function.
However, if the values of the parameters are known (or supposed)
to satisfy some constraint (which may or may not be true) then the
values of the parameters which minimise the potential energy func-
tion subject to the constraint will be associated with a value of this
function which is necessarily no smaller than the unconstrained
value. We may therefore use this difference in potential energy
levels as the basis of a natural test of the hypothesis that the values
of the parameters in the underlying statistical model truly satisfy
the given constraint. For example, in the context of the linear
regression models of Chapters 4 and 5, scaled variants of this
difference in potential energy levels define the Wald and
Lagrange multiplier statistics whereas the difference between the
logarithms of the potential energy levels defines the logarithm of
the conventional likelihood ratio statistic.
Figure 2.1 shows a bowl-shaped potential energy function
defined as a function of the parameters b and a where b increases
as we move to the right of the figure, and a increases as we move
16 Chapter 2
Figure 2.1 General potential energy function.
downwards and to the left. Also shown are some typical curves in
the surface of this function corresponding to successively larger
values of a. The overall (unconstrained) minimum of this function
corresponds to the lowermost point on the fifth of these curves. If,
however, the value of a is known (or supposed) to take the value
represented by the third curve, then the constrained minimum
corresponds to the lowermost point on this curve, and this optimal
point is necessarily associated with a larger value of the potential
energy function. If the Wald, Lagrange multiplier, or likelihood
ratio statistic defined by the difference between these two potential
energy levels is larger than an agreed critical value, then we would
be inclined to reject the hypothesis that a truly takes the value
implied by the third curve.
2.2 Geometrical Representations in

Prediction Space
Putting aside these abstract mechanical models for the present, we
turn our attention to geometrical representations of the data.
There are at least three distinct geometrical representations of a
set of observations; we shall refer to these three representations as
lying in the prediction, observation, and parameter spaces respec-
tively. The first two representations are more familiar to statisti-
cians than the third; however, apart from a brief mention in
Chapter 5, we shall have no role for the prediction space represen-

tation in this book. On the other hand, we shall make considerable
use of the little known parameter space representation in Chapters
5, 6, and 7.
2.2.1 Prediction Space Representation

Suppose that we have a set of n observations y1 ; y2 ; . . . ; yn on a
single variable Y, then what we shall name the prediction space
representation of this set of observations makes use of an n-dimen-
sional Cartesian space with n mutually orthogonal axes. The first
of these axes is associated with the value taken by the first obser-
vation, the second axis with the value taken by the second obser-
vation, and so on. Thus, in this representation, we identify the set
of n observations on the variable Y with the single point in this
space with coordinates ðy1 ; y2 ; . . . ; yn Þ relative to the given system
of orthogonal axes.
Now, suppose that we also have a set of n observations x1 ; x2 ;
. . . ; xn on a second variable X, then the prediction space represen-
tation of this second set of observations makes use of the same n-
dimensional space and represents this set of observations as a point
in this space with coordinates ðx1 ; x2 ; . . . ; xn Þ relative to the same
system of axes. If each of these n observations on X is multiplied by
the same factor b where 0 b 1; then the result may be identified
as a point on the line segment joining the origin ð0; 0; . . . ; 0Þ of the
space to the point with coordinates ðx1 ; x2 ; . . . ; xn Þ and lying a
proportion b of the distance from the origin of this point. If the
parameter b is permitted to range over all values in the interval 0
b 1 then, as indicated in Figure 2.2, this set of points will trace
the entire line segment between these two points. This line segment
is often used as a geometrical representation of the abstract vector
determined by this set of observations. (Note that we have marked
the axes of Figure 2.2 with the symbols X1 ; X2 ; and X3 when we
should perhaps have used the symbols X1 ; Y1 on the first axis,
X2 ; Y2 on the second, and X3 ; Y3 on the third, see Figure 2.3.
However, in general, we prefer to avoid this multiplicity of symbols
by adopting one of the relevant variables as an indicator. It would
18 Chapter 2
Figure 2.2 Prediction space representation of a set of observations.
perhaps have been more correct to adopt an imaginary variable as

indicator and to mark the axes E1 ; E2 ; and E3 :Þ
2.2.2 A Geometrical Analogy for the

Correlation Coefficient
Let O; X; and Y denote the points in n-dimensional prediction
space with coordinates (0, 0, . . ., 0), ðx1 ; x2 ; . . . ; xn Þ and
ðy1 ; y2 ; . . . ; yn Þ. Further, let represent the angle between the line
segments OX and OY, then it is well known that the squared
distance between the points X and Y is related to the squared
distance between the points O and X and the squared distance
between the points O and Y by the following generalisation of
Pythagoras’s theorem:
X X X pffiffi X X
ðxi yi Þ2 ¼ x2i þ y2i 2 ½ x2i y2i cosðÞ
Now, the expression on the left of this equation may be written in
the form
X X X X
ðxi yi Þ2 ¼ x2i þ y2i 2 xi yi
so that the cosine of the angle between the lines OX and OY
Figure 2.3 Prediction space representation of the correlation coefficient.
P
x yi
cosðÞ ¼ pffiffi P 2i P
½ x i y2i
is a geometrical model for the correlation between the variables X

and Y. The familiar formula for simple correlation
P
ðx x Þðyi yÞ
cosðÞ ¼ pffi P i P
½ ðxi x Þ2 ðyi yÞ2
is immediately obtained if the observations on X and Y are
assumed to be expressed in deviations from their arithmetic
means. See Leung and Lam (1975) for a detailed account of this
geometrical interpretation of the correlation coefficient, and
Rousseeuw and Molenberghs (1994) for more recent developments
in this area.
2.2.3 Planes and Hyperplanes in

Prediction Space
In this book we shall not be concerned with models suggested by
the finite line segment between the points O and X, but with those
suggested by the straight line of infinite length through these points
20 Chapter 2
which is traced out when the parameter b of Subsection 2.2.1 is

permitted to range over all values of b, that is when
1 < b < þ1.
Now suppose that we also have a set of observations on a
third variable Z, then this observation may be represented by a
point in the same n-dimensional space as the observations on X,
and all common multiples of the observations on Z may be repre-
sented as a second line through the origin of the same space. In a
similar way, the set of points determined by adding a multiple of
the observations on X to a (distinct) multiple of the corresponding
observations on Z determines a two-dimensional plane passing
through the origin and the points determined by the sets of obser-
vations on X and Z respectively.
In a similar way, if we have sets of n observations on q expla-
natory variables X1 ; X2 ; . . . ; Xq then these observations determine
a set of q points in n-dimensional space and, in general, the set of
all linear combinations of these q sets of observations determines a
q-dimensional plane or hyperplane passing through the q given
points and the origin. (In passing we note that some authors prefer
to reserve the word hyperplane for the special case in which the
Figure 2.4 Prediction space representation of linear combinations of two sets

of observations.
dimension of the subspace is one less than the dimension of the

whole space, that is when q ¼ n 1:)
We shall defer our discussion of the prediction space repre-
sentation of the data until after we have examined the relevant line
fitting procedures in Chapter 5. Even then, our account of the
subject will be brief and we refer readers to Durbin and Kendall
(1951) for further details. In this context, it should be noted that
they refer to our ‘‘prediction space’’ as ‘‘observation space.’’
2.3 Observation Space

Representation
In the previous section, we have seen that the prediction space
representation of the n observations on a set of q explanatory
variables and a single dependent variable is as a set of q þ 1 points
in n-dimensional space. By contrast, what we shall call the obser-
vation space representation of these n observations is as a set of n
points in ðq þ 1Þ-dimensional space. If this set of observations
exactly satisfies the relationship
yi ¼ xi1 b1 þ xi2 b2 þ . . . þ xiq bq i ¼ 1; 2; . . . ; n
for some sets of values b1 ; b2 ; . . . ; bq , then these n points will lie on
a q-dimensional hyperplane in this ðq þ 1Þ-dimensional space.
Further, if all the observations on one of the variables, say the
first, take the same value, say unity, then the dimensions of both
the space itself and the specified hyperplane may be reduced by
unity. Thus, if this set of observations exactly satisfies the relation-
ship
yi ¼ b1 þ xi2 b2 þ . . . þ xiq bq i ¼ 1; 2; . . . ; n
for some set of values b1 ; b2 ; . . . ; bq , then they will lie on a ðq 1Þ-
dimensional hyperplane in q-dimensional space.
Restricting our discussion to the case q ¼ 2 with xi1 ¼ 1, we
find that the ith observation X1 ¼ 1; X2 ¼ xi , and Y ¼ yi on the
variables X1 ; X2 , and Y may be represented by the point ðxi ; yi Þ in
the two-dimensional Cartesian plane with its x-axis running from
west to east along the line y ¼ 0 and its y-axis running from south
22 Chapter 2
to north along the line x ¼ 0. Further, we note that all observa-

tions satisfying the relationship
yi ¼ a þ bxi i ¼ 1; 2; . . . ; n
will lie on a (one-dimensional) straight line in this two-dimensional
plane. This situation is illustrated in Figure 2.5.
The point ðx; yÞ ¼ ð4; 6Þ is also plotted in this figure, for
reasons to be explained below.
2.4 Parameter Space

Representation
Again restricting our discussion to the case q ¼ 2 with xi1 ¼ 1, we
find that we may also represent the set of all observations satisfying
the relationship
yi ¼ a þ bxi i ¼ 1; 2; . . . ; n
by the values of the parameters a and b or, more formally, we may
represent this set of observations by the point (b, a) in the two-
dimensional Cartesian parameter plane with its b-axis running
from west to east along the line a ¼ 0 and its a-axis running
from south to north along the line b ¼ 0. Thus a line in the xy-
Figure 2.5 Observation space representation of a single observation and a

linear relationship between a set of observations.
plane corresponds to a point in the ba-plane. (Note that, for our

later convenience, we have reversed the lexicographical ordering of
the letters a and b here. However, there is no formal reason why we
cannot employ these letters in their usual ordering should we wish
to do so.)
This situation is illustrated in Figures 2.5 and 2.6 as the line
y ¼ 1 þ 2x (with parameters a ¼ 1 and b ¼ 2) in the xy-plane of
Figure 2.5 may be represented by the point ðb; aÞ ¼ ð2; 1Þ in the ba-
plane of Figure 2.6
Now, consider what is implied by the line
a ¼ y0 x0 b
in the ba-plane. It is clear that this equation defines the set of
intercepts a and slopes b of the family of lines that pass through
the point ðx0 ; y0 Þ in the xy-plane. This situation is illustrated in
Figure 2.7.
We note that the points (0.0,6), (0.5,4), (1.0,2), and (1.5,0) all
lie on the line a ¼ 6 4b in the ba-plane of Figure 2.7. The lines in
the xy-plane y ¼ 6; y ¼ 4 þ 0:5x; y ¼ 2 þ x; and y ¼ 1:5x corre-
sponding to these four points in the ba-plane all pass through the
point ðx; yÞ ¼ ð4; 6Þ in the xy-plane of Figure 2.8, and the rela-
tionship between the line a ¼ 6 4b in the ba-plane and the point
ðx; yÞ ¼ ð4; 6Þ in the xy-plane is established.
Figure 2.6 Parameter space representation of a linear relationship between

observations.
24 Chapter 2
Figure 2.7 Parameter space representation of a linear relationship between

parameters.
Thus, in the same way as a line in the xy-plane corresponds to

a point in the ba-plane, a line in the ba-plane will correspond to a
point in the xy-plane.
However, there is a subtle difference, as may be seen by com-
bining Figures 2.6 and 2.7 to form Figure 2.9 before comparing the
line a ¼ 6 4b and the point ðb; aÞ ¼ ð2; 1Þ in the ba-plane of
Figure 2.9 with the point ðx; yÞ ¼ ð4; 6Þ and the line y ¼ 1 þ 2x
in the xy-plane of Figure 2.5. Clearly, a point in the positive quad-
Figure 2.8 Observation space representation of a linear relationship between

parameters.
Figure 2.9 Parameter space representation of a set of parameters and a linear

relationship between sets of parameters.
rant of the xy-plane defines a line in the ba-plane with a negative

slope, whilst a point in the positive quadrant of the ba-plane
defines a line in the xy-plane with a positive slope. Indeed, the
particular form of this relationship establishes that the parameter
space representation of a fitting problem is the projective geometry
dual of the observation space representation of the same problem
and vice versa.
2.5 Line Fitting by Eye

As an exercise in the use of these last two representations, and as a
preparation for our later discussion of the line fitting problem, the
reader is invited to consider the problem of choosing a line which
seems best to fit the n ¼ 5 observations on X and Y given in Table
2.1. In the present context, this problem is to be solved without the
aid of an explicit optimality criterion.
As explained in Section 2.3, the artificial data given in Table
2.1 may be plotted as a scatter of points in the xy-plane, see Figure
2.10. Using the edge of a transparent ruler or a taut length of black
cotton, we may adjust the position of the straight line defined in
this way until it determines a line which seems to fit this set of
observations as closely as possible (where, for the time being, it is
26 Chapter 2
Table 2.1 Simple numerical example
X Y
2 2
3 6
4 4
6 8
8 4
left to the reader to decide what is meant by the phrase ‘‘as closely
as possible’’).
The results obtained here without the benefit of an explicit
optimality criterion may be compared with the lines fitted by two
variants of the method of least squares illustrated in Figures 1.1
and 1.2.
This traditional line fitting technique may readily be auto-
mated by using a computer mouse to adjust the positions of the
midpoint and one end of the candidate fitted line in the xy-plane.
This is the technique recommended by Bajgier, Atkinson, and
Prybutok (1989). However, their proposal involves the use of a
Figure 2.10 Observation space representation of the data in Table 2.1.

single computer mouse for two distinct purposes. It is more con-

venient to employ the mouse to identify suitable values for the
slope and intercept parameters of the candidate line directly in
the ba-plane. In principle, we have to display two planes on the
same monitor screen, one to represent the positions of the observed
points and the fitted line in observation space, and the other to
represent the values of the slope and intercept parameters of the
fitted line in parameter space. The desired result may either be
achieved by splitting the monitor screen into two portions, or by
employing a toggle to alternate between the two representations of
the problem.
We shall not develop a computer program of the type outlined
here. But, supposing that we had done so, then the practitioner
would have been asked to use the computer mouse in the ba-plane
to determine the position of the point corresponding to the slope
and intercept parameters of the candidate fitted line in the xy-
plane. The position of the line which seems best to fit the given
observations would then be determined by visual inspection.
At the conclusion of the fitting process we might have a pair of
screens such as those indicated in Figure 2.11.
This empirical determination of a supposedly optimal fitted
line suffers from the obvious defect that the practitioner’s choice is
made without reference to an explicit goodness of fit criterion. If it
is thought appropriate to employ the least squares criterion of
Chapter 4 in this role, then this criterion may readily be incorpo-
Figure 2.11 Dual representations of the line fitting problem.

Random documents with unrelated
content Scribd suggests to you:
Paroid Roofing
Partial View of Our Mills on the Neponset River at East Walpole,

Mass.
Paroid is Made from Start to Finish Right in Our Own Mills.
Our products are for the man who is planning new buildings, or
about to make repairs to old ones; and we have tried to tell in the
following paragraphs how each one of our materials is particularly
adapted to the different kinds of work for which it is made. Our
claims are based on actual experiences and if you are in the market
for roofing or sheathing papers, you will find that our materials will
save you money. First of all, we are going to tell you about our
Paroid Roofing, because the roof is one of the most important parts
of every building. If it is not right, there is no end of trouble.
THE DIFFICULTIES OF CHOOSING A GOOD ROOF. There
are about thirty different brands of ready roofing, and for most of
them the same claims are made. Under those conditions, how are
you going to choose the one that will prove most economical for
you? There is only one test that will tell, and that is the test of time.
Of course you can’t make that test yourself, but it is for your
interests to find out if others have made it and for how long.
The most economical roofing is not the one that costs you the
least per roll when you buy it, but the roofing that costs you least
per year of service. We are going to tell you here the most important
facts about Paroid; how it compares with shingles, metal and other
kinds of ready roofing, and then you can be your own judge.
PAROID vs. TIN AND IRON ROOFS. The best quality of tin,
iron and steel roofings cost much more than Paroid at the start, and
then there is always the additional cost of painting each year. In
spite of all you can do, a metal roof will rust out and spring leaks
which cannot be permanently repaired. Paroid cannot rust; it costs
less than metal roofs when you buy it, and less to apply. Anyone can
lay Paroid. One example that proves the superiority of Paroid over
metal roofs was shown when the Chicago & Northwestern Railroad,
who had tried different kinds of roofing on their Chicago train sheds,
including a good tin roof, used Paroid when the tin roof failed. The
Paroid Roofing is still in good condition.
PAROID vs. SHINGLES. If you have recently asked your
lumber dealer for a price on shingles, you are probably looking for a
substitute because of the exorbitant price asked for them. Lumber is
scarce everywhere, and shingles are growing poorer in quality and
higher in price every day. The test of time has proved that Paroid is
the real substitute for shingles, and it has many advantages which
shingles do not have. Figure this out for yourself. The first cost of
Paroid is less than that of a medium grade of shingles. You can lay
Paroid yourself and it requires an experienced man to lay shingles.
Shingles catch fire easily, while Paroid is practically fireproof against
sparks, cinders and embers.
Read what a large lumber dealer in Maine says about the
comparative cost of shingles and Paroid Roofing. He is right in the
heart of the shingle belt, and naturally the difference is not so great
as in other sections of the country where shingles are not so
plentiful.
Colony chicken houses on farm of G. M. Gowell, of the
Maine Agricultural Experiment Station, Orono, Maine.
The roofs are covered with PAROID, sides with NEPONSET.
See plans and description on page 17.
One-ply Paroid, which is usually heavy enough for the roof and
sides of most farm and poultry buildings, will save you at least 35%
over the cost of shingles. Here are the figures showing the
comparative cost of one-ply Paroid and B. C. Cedar Shingles.
Clear cedar shingles per square $3.10
4 lbs. nails at 3¢ per lb. .12
Average cost of carpenter labor 1.25 4.47
Cost per square foot 45¢
1-ply Paroid per square $2.50

Laying .35 2.85
Cost per square foot 28½¢
Lumber dealers all over the United States and Canada who
previously sold shingles exclusively, now sell large quantities of
Paroid Roofing. That tells the whole story.
In the core of each roll of Paroid sufficient nails, rust-proof caps,
cement, and complete directions for applying are packed. You can
lay it yourself with a hammer and knife. One-ply Paroid, costing
about one half as much as shingles, is heavy enough for most farm
and poultry buildings. For barns, stables, and other large buildings
we recommend two-ply, which is heavier and thicker.
PAROID vs. OTHER READY ROOFINGS
There are certain qualities that all ready roofings must have, but
the important question is, How long do they keep these qualities?
We have made our story short, but at the same time complete
enough, so that you can be your own judge when you compare our
claims for Paroid with the claims of other manufacturers.
A hog house on a Vermont farm, covered with PAROID.
READY ROOFING EXPERIENCE. You have probably read the

advertisements of some manufacturers who claim that because they
have had fifty to one hundred years’ experience (in some business or
other), that they make the best ready roofing. We have been making
felt, paper and roofing materials here in our own mills for nearly one
hundred years (established in 1817) but WE REFUSE TO CLAIM
that the length of time we have been established has anything to do
with the real merit of Paroid Roofing. It shows only that we have had
the right kind of experience. We maintain that the test of time is the
only real test of a roofing. Paroid has stood this test.
Paroid covers the Plant Industry Buildings,
U. S. Department of Agriculture,
Washington, D. C.
THE OLD CRY OF THE IMITATORS. Most manufacturers warn

you against “imitations” of their “genuine” ready roofing.
Investigate, and you’ll find that the imitators themselves are the first
to talk about imitations. The question of imitations has nothing to do
with the merits of a particular roofing. Some imitations are often
better than the originals; but there is only one way to prove it—the
test of time is the test that tells. Paroid has stood this test.
PAROID IS MADE BETTER THAN OTHER READY
ROOFINGS. There are three important things that enter into the
manufacture of ready roofing; namely, the felt, saturation and the
coating.
We make the felt for Paroid Roofing in our own mills because we
could not entrust to others the making of the most important part of
Paroid. If the felt is not right the roofing will not be. Do not run
risks. Buy your roofing from manufacturers who make their own felt.
SATURATION AND COATING. The strong, well-made Paroid
Roofing felt is soaked, not merely dipped, in a compound of our
own, rendering every fibre of it absolutely proof against water, cold
and heat. The felt is then given a thicker, smoother and more pliable
coating than that on any other ready roofing. Compare samples and
you will see and feel the difference. It is more flexible in cold, and it
will not melt or run in the heat. It is more sightly, and lays easier
and smoother than any other ready roofing.
Roof of Brooder house, covered with PAROID,

White Leghorn Poultry Yards, Waterville, N. Y.
See plan and description on page 12.
WE WERE THE ORIGINATORS OF THE

COMPLETE ROOFING KIT
Inside of each roll of Paroid is packed cement, nails, rust-proof
caps, and complete directions for applying. Anyone can lay Paroid
and get good results if the directions are carefully followed.
OUR PATENTED RUST RETARDING CAPS

Paroid is the only ready roofing supplied with rust retarding caps
for applying the roofing to a building. They are square, and
therefore, have more binding surface than the ordinary round caps.
The nails are also coated with a rust retarding preparation.
Largest stock barn on the largest stock farm in Minnesota.
Eight hundred squares of PAROID put on roof of this barn by the
farm hands themselves.
PAROID FOR FARM AND POULTRY BUILDINGS

Paroid is adapted to all kinds of buildings and especially farm and
poultry buildings. It makes a building warmer in winter and cooler in
summer than other kinds of roofing. It will not taint rain water and is
not affected by gases and fumes.
PAROID FOR SIDING

The next time you put up a poultry house, shed, or other farm
building, lay Paroid on the roof yourself, and then apply it to the
sides with battens. You will be surprised at the neat effect it gives,
and it is more economical than clapboards and shingles.
OUR GUARANTEE
You run no risks when you buy Paroid. Every roll is sold on this
guarantee.
Buy a roll of Paroid; open it; examine it; apply it to your roof;
and then, if you are not satisfied that you have the best ready
roofing on the market, send us your name and address and we will
send you a check for the full amount you have paid for the roofing,
including the cost of applying it.
D. J. Lambert says, “PAROID is all right;” and he knows.
Our dealers all over the country who handle Paroid will also make
you this offer. If your dealer does not carry Paroid in stock, send us
your order and check or money order direct. We will pay the freight.
Bird’s-eye View of Egg Plant.
W. Harry Owen’s Farm, Vineyard Haven, Mass.
All buildings are covered with PAROID.
PRICES
You can pay most any price for a ready roofing, and, like
everything else, you get as much quality as you pay for. Paroid may
cost more than other ready roofings the day you buy it, but it is less
expensive after it is applied to your roof, because it will last longer.
We maintain that the test of time is the test that tells, and Paroid
has stood that test. Don’t make a mistake and buy a roofing that will
go to pieces in a short time. If you would save money, choose the
roofing that lasts the longest.
PRICE LIST
Paroid Roofing, 1-ply, $2.50 per square (100 sq. ft.).
Paroid Roofing, 2-ply, $3.50 per square (100 sq. ft.).
These prices include extra roofing for laps and nails, rust-proof
caps, cement, and directions for applying.
Meadow Brook Farm poultry plant, Dallas, Pa.

Roofs are covered with PAROID.
Monmouth Poultry Farm, Freneau, N. J. Roof covered with

PAROID.
Amboy, Ill., Dec. 29, 1905.

F. W. Bird & Son, Chicago, Ill.
Please send me your up-to-date poultry and farm
building plans. I use Paroid, and can find no equal. I
am going to build a hog house, and will cover it with
Paroid.
Yours respectfully,
F. M. Blowers.
Aug. 15, 1905.

Messrs. F. W. Bird & Son, East Walpole, Mass.
Gentlemen: Of three kinds of paper used this
season on my bee hive covers, your Paroid has given
such satisfaction that I would use nothing else in
future.
Yours respectfully,
E. H. Dewey.
Great Barrington, Mass.
Akron, Ohio, June 11, 1906.

F. W. Bird & Son, Chicago, Ill.
Gentlemen: Enclosed you will find the slip that I
received from you filled out with my address and a
two-cent stamp, for which please send me the book
“Practical Farm Buildings,” and oblige.
In regard to your Paroid Roofing will say that it is
the best that I have seen to date, for durability,
easiness to lay, and its wearing qualities.
I have used it on my poultry buildings for the last
ten years, and if occasion demands that I shall need
any more roofing for any additional poultry buildings
PAROID it shall be.
So hoping to receive the book as soon as
convenient for you to send it, I remain,
Yours respectfully,
Paul C. Bork.
343 Hickory St.
PAROID covers a Vermont barn and silo.

The Largest Duck Farm in the World.
Duck breeding house, roof and sides covered with PAROID,
Weber Bros., Pondville, Mass.
Willow Brook Farm, Berlin, Conn.
All poultry buildings are covered with PAROID.
See testimonial below.
Berlin, Conn., Jan. 24, 1906.

F. W. Bird & Son,
East Walpole, Mass.
Dear Sirs: We find your Paroid roofing paper the
very best we have ever used. We have thousands of
visitors who are looking for information in regard to
roofing paper each year, and in each and every
instance we recommend your roofing paper. We have
done this because we think it is the best out, and will
take pleasure in recommending it in the future.
Very truly yours,
Willow Brook Farm.
Pigeonry on Jordan’s Hackney Stud Farm, Plymouth, Mass.

Covered with PAROID.
Round House of the New York Central & Hudson River Railroad,
near High Bridge, N. Y.
City Detroit, Mich.,

of
Department of Public Works,
March 28, 1906.
F. W. Bird & Son,
East Walpole, Mass.
Gentlemen: Please send me a copy of your book of
farm and poultry building plans, and oblige. I have one
building covered with your Paroid, and like it very
much. It wears well and gives no trouble.
Yours truly,
(Signed) Porter Murphy
.
82 Perry St., Detroit, Mich.
Tongue Point Lumber Co., Astoria, Oregon.
Covered with PAROID.
Woodbury & Walker Block, Burlington, Vermont.

Roofed with PAROID.
Neponset Red Rope
Roofing
For over twenty-five years Neponset has been the standard low
cost roofing and siding. It must not be compared with tarred felts
just because it costs about the same. Neponset will usually outlast
them three to one.
Neponset won’t run and dry out like tarred felts. It is easier to
apply and cleaner to handle.
Figure it out for yourself. A tarred felt costing the same as
Neponset lasts only a few seasons. Neponset lasts at least from five
to seven years and in most cases longer. If you are going to use a
low cost roofing, Neponset will save you money.
Neponset makes a practically permanent siding, and if Paroid is
too expensive for both roof and sides, we recommend Paroid for the
roof and Neponset for the sides.
Neponset is put up in rolls 36 inches wide, containing, 100, 250
and 500 square feet. Fixtures and directions for applying Neponset
are packed inside of each roll.
NEPONSET BLACK WATERPROOF PAPER is made especially
for sheathing purposes, but it will last a year or two on the roof or
sides of buildings. It costs less and is cleaner to handle than tarred
felts. Neponset Black is put up in rolls 36 inches wide containing 250
and 500 square feet.
NEPONSET WATERPROOF SHEATHING PAPER

One of the most important items that every house builder ought
to consider is that of sheathing papers. This important question,
unless decided right, means an additional expense of many dollars in
fuel each year. Sheathing papers are used to keep out cold and
dampness, but only a few fulfill their purpose. Cold draughts
penetrate cheap papers, and in a very short time these cheap papers
disintegrate and become mere dust. A good waterproof paper repels
dampness, keeps out the cold and lasts the life of a building. For
over twenty-five years Neponset Papers have been the standard.
Actual experiences have proved that Neponset saves one-third of the
fuel required to heat a house, therefore, a big saving each year.
Neponset acts as a blanket on a house it keeps out the cold and
keeps in the heat. Don’t lay the foundations for an annual loss, save
one-third of the money you would spend on fuel. That’s what
Neponset has done for others, it will do it for you.
Bird’s-eye View of Chas. F. Thompson & Co.’s Poultry Plant,

Lynnfield Centre, Mass.
See testimonial.
Lynnfield Centre, Mass.,
April 3, 1906.
Messrs. F. W. Bird & Son,
East Walpole, Mass.
Gentlemen: Replying to yours of the 2nd, there is
no photographer here that can take views. I am
sending you a catalogue showing views we have half-
tone plates for. If they will do I can loan them to you.
The original photographs are lost. The long buildings
shown, bird’s-eye view are covered with Neponset, put
on nearly ten years ago; one coat of paint put on at
the time, nothing done since and not a leak; appear in
good condition now. Anything we can do for you let us
know.
Chas. F. Thompson & Co.
Newburgh, N. Y.,
April 13th, 1906.
F. W. Bird & Son,
East Walpole, Mass.
Gentlemen: Your favor at hand. Samples of leaflets only had
Neponset on. Hope that you will also send some with Paroid as we
find, in many cases, customers prefer to pay the difference.
One of our roofs, about twelve hundred square feet, was covered
with Red Rope some ten years ago. It had no care, but kept OK.
until this season. Pretty good record for a cheap roof.
Yours very truly,
The Newburgh Lumber Co.

A Pennsylvania Duck Farm.
NEPONSET is especially adapted for poultry houses.
Kansas City,
Nov. 24, 1905.
The Kansas City Roofing & Corrugating Co.,
Kansas City, Mo.
Gentlemen: Replying to yours of the 22nd inst., we covered our
large lumber shed and barn with Neponset Red Rope Roofing two
years ago this fall, and painted same at once, and to this day is in
good condition and has given us entire satisfaction. Other kinds of
roofing which we paid more money for do not seem to have given us
the service that this has. We ask you to kindly advise us what is the
best paint to repaint this with, and which is the best season of the
year to use same.
Yours very truly,
Badger Lumber Co.,

(Signed) L. J. Gilles, Agent.
Wayland, N. Y.,
Oct. 20, 1906.
F. W. Bird & Son,
East Walpole, Mass.
Gentlemen: Will you please send me sample of sidings and
roofings, and also prices? I want to get Neponset unless you have
got something better. Neponset beats anything of the kind I ever
used.
Yours truly,
W. E. Moulton.
Dancing Pavilion, Easton, Pa.

Roof is covered with NEPONSET.
Proslate Roofing and
Siding
PATENTED MAY 13, 1906
DESIGNED ESPECIALLY FOR PITCH ROOF HOUSES AND
FOR A PERMANENT SIDING
SPECIAL FEATURES
DESCRIPTION. Proslate is our regular Paroid roofing material
with an additional wearing surface (not a colored coating) of a
mixture of paint and sand, making a slate-like surface, slate in color
and effect. Proslate is an entirely new material—patented.
FOR PITCH ROOFS. Proslate is especially designed to take the
place of shingles and clapboards for residences. Attractiveness and
economy are both secured by using Proslate for pitch roofs. It is
finished in 18-inch rolls, ready to lay, with ornamented edge.
Proslate is applied in the usual way—lapped, cemented and
nailed—no waste by excessive overlapping—cement and fixtures of
same slate color with complete directions for laying are packed in
each roll. Any good carpenter can apply it.
Proslate is finished in rolls containing 122 square feet, sufficient
to cover 100 square feet of surface and is sold on a basis of material
enough to cover 100 square feet.
FOR FLAT ROOFS AND SIDING. For flat roofs and as a siding,
we furnish Proslate in rolls 36 inches wide, plain straight edges. By
the use of broad cleats, a very neat effect can be made on the sides
of houses. Proslate for a siding is warmer than clapboards or
shingles. It acts as a blanket.
GENERAL USE. As a permanent all round roof, we believe that
Proslate represents the best material for the money yet made. It is
good enough for residences, factories and railroad buildings.
Shingles are unsatisfactory and cannot be used on flat roofs,
porches, etc., and slate is too expensive. Proslate fills every
requirement of a good roofing, well made with an extra weather
surface, economical—permanent.
Florian Sound-Deadening Felt
This material is used between floors as an insulator against
sound. It is, without any exception, the cleanest and most effective
sound-deadening felt made. The corrugations make small dead air
cells when the felt is placed in position and this is considered to be
the most effective method of deadening sound. Tests have proved
that one sheet of Florian is equal to six sheets of ordinary deadening
felt. Florian is also a good non-conductor of fire, heat and cold. It
should be used between floors in every house to insure a well
insulated building.
OTHER THINGS WE MAKE

We make Neponset and Kosat Insulating Papers for cold storage
work; Parine Paint, especially for our own roofings, and all kinds of
outside work; Tack, Screw, and Shoe boxes, all kinds of special
papers and paper boxes.
We have had a large experience with all kinds of building and
roofing construction, and if we can help you on any of your problems
please be sure to write us.
F. W. BIRD & SON, MAKERS
ESTABLISHED 1817
EAST WALPOLE, MASS.

NEW YORK WASHINGTON CHICAGO
CANADIAN FACTORY AND OFFICE, HAMILTON, ONT.

WESTERN CANADIAN OFFICE, WINNIPEG, MAN.
Transcriber’s Notes:
The illustrations have been moved so that they do not break up paragraphs and so
that they are next to the text they illustrate.
Typographical and punctuation errors have been silently corrected.
*** END OF THE PROJECT GUTENBERG EBOOK PRACTICAL FARM
BUILDINGS: PLANS AND SUGGESTIONS ***
Updated editions will replace the previous one—the old editions will
be renamed.
Creating the works from print editions not protected by U.S.

copyright law means that no one owns a United States copyright in
these works, so the Foundation (and you!) can copy and distribute it
in the United States without permission and without paying
copyright royalties. Special rules, set forth in the General Terms of
Use part of this license, apply to copying and distributing Project
Gutenberg™ electronic works to protect the PROJECT GUTENBERG™
concept and trademark. Project Gutenberg is a registered trademark,
and may not be used if you charge for an eBook, except by following
the terms of the trademark license, including paying royalties for use
of the Project Gutenberg trademark. If you do not charge anything
for copies of this eBook, complying with the trademark license is
very easy. You may use this eBook for nearly any purpose such as
creation of derivative works, reports, performances and research.
Project Gutenberg eBooks may be modified and printed and given
away—you may do practically ANYTHING in the United States with
eBooks not protected by U.S. copyright law. Redistribution is subject
to the trademark license, especially commercial redistribution.
START: FULL LICENSE

THE FULL PROJECT GUTENBERG LICENSE
PLEASE READ THIS BEFORE YOU DISTRIBUTE OR USE THIS WORK
To protect the Project Gutenberg™ mission of promoting the free

distribution of electronic works, by using or distributing this work (or
any other work associated in any way with the phrase “Project
Gutenberg”), you agree to comply with all the terms of the Full
Project Gutenberg™ License available with this file or online at
www.gutenberg.org/license.
Section 1. General Terms of Use and

Redistributing Project Gutenberg™
electronic works
1.A. By reading or using any part of this Project Gutenberg™
electronic work, you indicate that you have read, understand, agree
to and accept all the terms of this license and intellectual property
(trademark/copyright) agreement. If you do not agree to abide by all
the terms of this agreement, you must cease using and return or
destroy all copies of Project Gutenberg™ electronic works in your
possession. If you paid a fee for obtaining a copy of or access to a
Project Gutenberg™ electronic work and you do not agree to be
bound by the terms of this agreement, you may obtain a refund
from the person or entity to whom you paid the fee as set forth in
paragraph 1.E.8.
1.B. “Project Gutenberg” is a registered trademark. It may only be

used on or associated in any way with an electronic work by people
who agree to be bound by the terms of this agreement. There are a
few things that you can do with most Project Gutenberg™ electronic
works even without complying with the full terms of this agreement.
See paragraph 1.C below. There are a lot of things you can do with
Project Gutenberg™ electronic works if you follow the terms of this
agreement and help preserve free future access to Project
Gutenberg™ electronic works. See paragraph 1.E below.
1.C. The Project Gutenberg Literary Archive Foundation (“the
Foundation” or PGLAF), owns a compilation copyright in the
collection of Project Gutenberg™ electronic works. Nearly all the
individual works in the collection are in the public domain in the
United States. If an individual work is unprotected by copyright law
in the United States and you are located in the United States, we do
not claim a right to prevent you from copying, distributing,
performing, displaying or creating derivative works based on the
work as long as all references to Project Gutenberg are removed. Of
course, we hope that you will support the Project Gutenberg™
mission of promoting free access to electronic works by freely
sharing Project Gutenberg™ works in compliance with the terms of
this agreement for keeping the Project Gutenberg™ name associated
with the work. You can easily comply with the terms of this
agreement by keeping this work in the same format with its attached
full Project Gutenberg™ License when you share it without charge
with others.
1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside the
United States, check the laws of your country in addition to the
terms of this agreement before downloading, copying, displaying,
performing, distributing or creating derivative works based on this
work or any other Project Gutenberg™ work. The Foundation makes
no representations concerning the copyright status of any work in
any country other than the United States.
1.E. Unless you have removed all references to Project Gutenberg:
1.E.1. The following sentence, with active links to, or other

immediate access to, the full Project Gutenberg™ License must
appear prominently whenever any copy of a Project Gutenberg™
work (any work on which the phrase “Project Gutenberg” appears,
or with which the phrase “Project Gutenberg” is associated) is
accessed, displayed, performed, viewed, copied or distributed:
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade
Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.
Let us accompany you on the journey of exploring knowledge and

personal growth!
ebookgate.com

45428

Uploaded by

Copyright:

Available Formats

45428

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

45428

Uploaded by

Copyright:

Available Formats

Visit https://ebookgate.

com to download the full version and

Visualizing Statistical Models and Concepts 1st

_____ Click the link below to download _____

Explore and download more ebooks at ebookgate.com

Visualizing Landscape Architecture Functions Concepts

Interpreting and Visualizing Regression Models Using Stata

Statistical Models 1st Edition A.C. Davison

Visualizing Project Management Models and Frameworks for

Statistical Misconceptions 1st Edition Schuyler W. Huck

Applied Linear Statistical Models 5th Edition Michael H.

Models for Probability and Statistical Inference Theory

Models for Probability and Statistical Inference Theory

Illustrated by Michaël Schyns

Marcel Dekker, Inc. New York • Basel

Copyright © 2002 by Marcel Dekker, Inc. All Rights Reserved.

This book is printed on acid-free paper.

Eastern Hemisphere Distribution

World Wide Web

Copyright # 2002 by Marcel Dekker, Inc. All Rights Reserved.

Current printing (last digit):

PRINTED IN THE UNITED STATES OF AMERICA

Statistical Computing/ Multivariate Analysis

Probability Quality Control/Reliability

Additional Volumes in Preparation

Visualizing Statistical Models and Concepts, R. W. Farebrother and Michael Schyns

Until relatively recent times, geometrical and mechanical analogies

lute deviations, the median absolute deviation, or the maximum

1.1 Introductory Remarks 1

Abstract Geometrical and Mechanical Representations 13

2.1 Bayesian and non-Bayesian Likelihood Functions 13

Mechanical Models for Multidimensional Medians 29

3.1 Mechanical Models for One-Dimensional Medians 29

3.3.2 Regional Constraints 49

Method of Least Squared Deviations 63

4.1 One- and Two-Dimensional Means 63

4.5.3 Measures of Inﬂuence 87

Method of Least Absolute Deviation 97

5.1 One- and Two-Dimensional Medians 97

5.5.2 Elemental Set Characterisation of the Weighted Least

Minimax Absolute Deviation Method 143

6.1 One- and Two-Dimensional Means 143

Method of Least Median of Squared Deviations 163

7.1 One- and Two-Dimensional Means 163

Mechanical Models for Metric Graphs 177

8.1 Introduction 177

Categorical Data Analysis 191

9.1 Nature of Categorical Data 191

Method of Averages and Curve Fitting by Splines 207

10.1 Mechanical Models for Multivariate Means 207

Multivariate Generalisations of the Method of Least

11.1 Multivariate Statistical Analysis 219

List of Figures 233

List of Tables 241

Name Index 243

Subject Index 245

1.1 Introductory Remarks

Figure 1.1 Side-by-side bar chart.

the observed points to the ﬁtted line as if they were springs or

Figure 1.2 End-on-end bar chart.

_ Click the link below to download _