Download the full version of the ebook at ebookfinal.

com

Statistical Analysis of Spatial and Spatio

Temporal Point Patterns Third Edition Peter J.
Diggle

https://ebookfinal.com/download/statistical-analysis-of-
spatial-and-spatio-temporal-point-patterns-third-edition-
peter-j-diggle/

OR CLICK BUTTON

DOWNLOAD EBOOK

Download more ebook instantly today at https://ebookfinal.com

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

Spatial Statistics and Spatio Temporal Data Covariance

Functions and Directional Properties 1st Edition Michael
Sherman
https://ebookfinal.com/download/spatial-statistics-and-spatio-
temporal-data-covariance-functions-and-directional-properties-1st-
edition-michael-sherman/
ebookfinal.com

Statistical Methods for Spatial Data Analysis 1st Edition

Oliver Schabenberger

https://ebookfinal.com/download/statistical-methods-for-spatial-data-
analysis-1st-edition-oliver-schabenberger/

ebookfinal.com

Parametric Statistical Change Point Analysis With

Applications to Genetics Medicine and Finance 2nd Edition
Jie Chen
https://ebookfinal.com/download/parametric-statistical-change-point-
analysis-with-applications-to-genetics-medicine-and-finance-2nd-
edition-jie-chen/
ebookfinal.com

Parametric Statistical Change Point Analysis With

Applications to Genetics Medicine and Finance 2nd Edition
Jie Chen
https://ebookfinal.com/download/parametric-statistical-change-point-
analysis-with-applications-to-genetics-medicine-and-finance-2nd-
edition-jie-chen-2/
ebookfinal.com
Spatial Analysis in Epidemiology Stevenson

https://ebookfinal.com/download/spatial-analysis-in-epidemiology-
stevenson/

ebookfinal.com

Reproduction in Cattle Third Edition Peter J. H. Ball

https://ebookfinal.com/download/reproduction-in-cattle-third-edition-
peter-j-h-ball/

ebookfinal.com

Measuring Customer Satisfaction and Loyalty Third Edition

Survey Design Use and Statistical Analysis Methods Bob E.
Hayes
https://ebookfinal.com/download/measuring-customer-satisfaction-and-
loyalty-third-edition-survey-design-use-and-statistical-analysis-
methods-bob-e-hayes/
ebookfinal.com

Statistical Methods in Spatial Epidemiology Second Edition

Andrew B. Lawson(Auth.)

https://ebookfinal.com/download/statistical-methods-in-spatial-
epidemiology-second-edition-andrew-b-lawsonauth/

ebookfinal.com

An Introduction to Statistical Problem Solving in

Geography Third Edition J. Chapman Mcgrew Jr.

https://ebookfinal.com/download/an-introduction-to-statistical-
problem-solving-in-geography-third-edition-j-chapman-mcgrew-jr/

ebookfinal.com
Statistical Analysis
of Spatial and Spatio-
Temporal Point
Patterns
Third Edition
 MONOGRAPHS ON STATISTICS AND APPLIED PROBABILITY

General Editors

F. Bunea, V. Isham, N. Keiding, T. Louis, R. L. Smith, and H. Tong

1. Stochastic Population Models in Ecology and Epidemiology M.S. Barlett (1960)

2. Queues D.R. Cox and W.L. Smith (1961)
3. Monte Carlo Methods J.M. Hammersley and D.C. Handscomb (1964)
4. The Statistical Analysis of Series of Events D.R. Cox and P.A.W. Lewis (1966)
5. Population Genetics W.J. Ewens (1969)
6. Probability, Statistics and Time M.S. Barlett (1975)
7. Statistical Inference S.D. Silvey (1975)
8. The Analysis of Contingency Tables B.S. Everitt (1977)
9. Multivariate Analysis in Behavioural Research A.E. Maxwell (1977)
10. Stochastic Abundance Models S. Engen (1978)
11. Some Basic Theory for Statistical Inference E.J.G. Pitman (1979)
12. Point Processes D.R. Cox and V. Isham (1980)
13. Identification of Outliers D.M. Hawkins (1980)
14. Optimal Design S.D. Silvey (1980)
15. Finite Mixture Distributions B.S. Everitt and D.J. Hand (1981)
16. Classification A.D. Gordon (1981)
17. Distribution-Free Statistical Methods, 2nd edition J.S. Maritz (1995)
18. Residuals and Influence in Regression R.D. Cook and S. Weisberg (1982)
19. Applications of Queueing Theory, 2nd edition G.F. Newell (1982)
20. Risk Theory, 3rd edition R.E. Beard, T. Pentikäinen and E. Pesonen (1984)
21. Analysis of Survival Data D.R. Cox and D. Oakes (1984)
22. An Introduction to Latent Variable Models B.S. Everitt (1984)
23. Bandit Problems D.A. Berry and B. Fristedt (1985)
24. Stochastic Modelling and Control M.H.A. Davis and R. Vinter (1985)
25. The Statistical Analysis of Composition Data J. Aitchison (1986)
26. Density Estimation for Statistics and Data Analysis B.W. Silverman (1986)
27. Regression Analysis with Applications G.B. Wetherill (1986)
28. Sequential Methods in Statistics, 3rd edition G.B. Wetherill and K.D. Glazebrook (1986)
29. Tensor Methods in Statistics P. McCullagh (1987)
30. Transformation and Weighting in Regression R.J. Carroll and D. Ruppert (1988)
31. Asymptotic Techniques for Use in Statistics O.E. Bandorff-Nielsen and D.R. Cox (1989)
32. Analysis of Binary Data, 2nd edition D.R. Cox and E.J. Snell (1989)
33. Analysis of Infectious Disease Data N.G. Becker (1989)
34. Design and Analysis of Cross-Over Trials B. Jones and M.G. Kenward (1989)
35. Empirical Bayes Methods, 2nd edition J.S. Maritz and T. Lwin (1989)
36. Symmetric Multivariate and Related Distributions K.T. Fang, S. Kotz and K.W. Ng (1990)
37. Generalized Linear Models, 2nd edition P. McCullagh and J.A. Nelder (1989)
38. Cyclic and Computer Generated Designs, 2nd edition J.A. John and E.R. Williams (1995)
39. Analog Estimation Methods in Econometrics C.F. Manski (1988)
40. Subset Selection in Regression A.J. Miller (1990)
41. Analysis of Repeated Measures M.J. Crowder and D.J. Hand (1990)
42. Statistical Reasoning with Imprecise Probabilities P. Walley (1991)
43. Generalized Additive Models T.J. Hastie and R.J. Tibshirani (1990)
44. Inspection Errors for Attributes in Quality Control N.L. Johnson, S. Kotz and X. Wu (1991)
45. The Analysis of Contingency Tables, 2nd edition B.S. Everitt (1992)
46. The Analysis of Quantal Response Data B.J.T. Morgan (1992)
47. Longitudinal Data with Serial Correlation—A State-Space Approach R.H. Jones (1993)
48. Differential Geometry and Statistics M.K. Murray and J.W. Rice (1993)
49. Markov Models and Optimization M.H.A. Davis (1993)
50. Networks and Chaos—Statistical and Probabilistic Aspects
O.E. Barndorff-Nielsen, J.L. Jensen and W.S. Kendall (1993)
51. Number-Theoretic Methods in Statistics K.-T. Fang and Y. Wang (1994)
52. Inference and Asymptotics O.E. Barndorff-Nielsen and D.R. Cox (1994)
53. Practical Risk Theory for Actuaries C.D. Daykin, T. Pentikäinen and M. Pesonen (1994)
54. Biplots J.C. Gower and D.J. Hand (1996)
55. Predictive Inference—An Introduction S. Geisser (1993)
56. Model-Free Curve Estimation M.E. Tarter and M.D. Lock (1993)
57. An Introduction to the Bootstrap B. Efron and R.J. Tibshirani (1993)
58. Nonparametric Regression and Generalized Linear Models P.J. Green and B.W. Silverman (1994)
59. Multidimensional Scaling T.F. Cox and M.A.A. Cox (1994)
60. Kernel Smoothing M.P. Wand and M.C. Jones (1995)
61. Statistics for Long Memory Processes J. Beran (1995)
62. Nonlinear Models for Repeated Measurement Data M. Davidian and D.M. Giltinan (1995)
63. Measurement Error in Nonlinear Models R.J. Carroll, D. Rupert and L.A. Stefanski (1995)
64. Analyzing and Modeling Rank Data J.J. Marden (1995)
65. Time Series Models—In Econometrics, Finance and Other Fields
D.R. Cox, D.V. Hinkley and O.E. Barndorff-Nielsen (1996)
66. Local Polynomial Modeling and its Applications J. Fan and I. Gijbels (1996)
67. Multivariate Dependencies—Models, Analysis and Interpretation D.R. Cox and N. Wermuth (1996)
68. Statistical Inference—Based on the Likelihood A. Azzalini (1996)
69. Bayes and Empirical Bayes Methods for Data Analysis B.P. Carlin and T.A Louis (1996)
70. Hidden Markov and Other Models for Discrete-Valued Time Series I.L. MacDonald and W. Zucchini (1997)
71. Statistical Evidence—A Likelihood Paradigm R. Royall (1997)
72. Analysis of Incomplete Multivariate Data J.L. Schafer (1997)
73. Multivariate Models and Dependence Concepts H. Joe (1997)
74. Theory of Sample Surveys M.E. Thompson (1997)
75. Retrial Queues G. Falin and J.G.C. Templeton (1997)
76. Theory of Dispersion Models B. Jørgensen (1997)
77. Mixed Poisson Processes J. Grandell (1997)
78. Variance Components Estimation—Mixed Models, Methodologies and Applications P.S.R.S. Rao (1997)
79. Bayesian Methods for Finite Population Sampling G. Meeden and M. Ghosh (1997)
80. Stochastic Geometry—Likelihood and computation
O.E. Barndorff-Nielsen, W.S. Kendall and M.N.M. van Lieshout (1998)
81. Computer-Assisted Analysis of Mixtures and Applications—Meta-Analysis, Disease Mapping and Others
D. Böhning (1999)
82. Classification, 2nd edition A.D. Gordon (1999)
83. Semimartingales and their Statistical Inference B.L.S. Prakasa Rao (1999)
84. Statistical Aspects of BSE and vCJD—Models for Epidemics C.A. Donnelly and N.M. Ferguson (1999)
85. Set-Indexed Martingales G. Ivanoff and E. Merzbach (2000)
86. The Theory of the Design of Experiments D.R. Cox and N. Reid (2000)
87. Complex Stochastic Systems O.E. Barndorff-Nielsen, D.R. Cox and C. Klüppelberg (2001)
88. Multidimensional Scaling, 2nd edition T.F. Cox and M.A.A. Cox (2001)
89. Algebraic Statistics—Computational Commutative Algebra in Statistics
G. Pistone, E. Riccomagno and H.P. Wynn (2001)
90. Analysis of Time Series Structure—SSA and Related Techniques
N. Golyandina, V. Nekrutkin and A.A. Zhigljavsky (2001)
91. Subjective Probability Models for Lifetimes Fabio Spizzichino (2001)
92. Empirical Likelihood Art B. Owen (2001)
93. Statistics in the 21st Century Adrian E. Raftery, Martin A. Tanner, and Martin T. Wells (2001)
94. Accelerated Life Models: Modeling and Statistical Analysis
Vilijandas Bagdonavicius and Mikhail Nikulin (2001)
95. Subset Selection in Regression, Second Edition Alan Miller (2002)
96. Topics in Modelling of Clustered Data Marc Aerts, Helena Geys, Geert Molenberghs, and Louise M. Ryan (2002)
97. Components of Variance D.R. Cox and P.J. Solomon (2002)
98. Design and Analysis of Cross-Over Trials, 2nd Edition Byron Jones and Michael G. Kenward (2003)
99. Extreme Values in Finance, Telecommunications, and the Environment
Bärbel Finkenstädt and Holger Rootzén (2003)
100. Statistical Inference and Simulation for Spatial Point Processes
Jesper Møller and Rasmus Plenge Waagepetersen (2004)
101. Hierarchical Modeling and Analysis for Spatial Data
Sudipto Banerjee, Bradley P. Carlin, and Alan E. Gelfand (2004)
102. Diagnostic Checks in Time Series Wai Keung Li (2004)
103. Stereology for Statisticians Adrian Baddeley and Eva B. Vedel Jensen (2004)
104. Gaussian Markov Random Fields: Theory and Applications Håvard Rue and Leonhard Held (2005)
105. Measurement Error in Nonlinear Models: A Modern Perspective, Second Edition
Raymond J. Carroll, David Ruppert, Leonard A. Stefanski, and Ciprian M. Crainiceanu (2006)
106. Generalized Linear Models with Random Effects: Unified Analysis via H-likelihood
Youngjo Lee, John A. Nelder, and Yudi Pawitan (2006)
107. Statistical Methods for Spatio-Temporal Systems
Bärbel Finkenstädt, Leonhard Held, and Valerie Isham (2007)
108. Nonlinear Time Series: Semiparametric and Nonparametric Methods Jiti Gao (2007)
109. Missing Data in Longitudinal Studies: Strategies for Bayesian Modeling and Sensitivity Analysis
Michael J. Daniels and Joseph W. Hogan (2008)
110. Hidden Markov Models for Time Series: An Introduction Using R
Walter Zucchini and Iain L. MacDonald (2009)
111. ROC Curves for Continuous Data Wojtek J. Krzanowski and David J. Hand (2009)
112. Antedependence Models for Longitudinal Data Dale L. Zimmerman and Vicente A. Núñez-Antón (2009)
113. Mixed Effects Models for Complex Data Lang Wu (2010)
114. Intoduction to Time Series Modeling Genshiro Kitagawa (2010)
115. Expansions and Asymptotics for Statistics Christopher G. Small (2010)
116. Statistical Inference: An Integrated Bayesian/Likelihood Approach Murray Aitkin (2010)
117. Circular and Linear Regression: Fitting Circles and Lines by Least Squares Nikolai Chernov (2010)
118. Simultaneous Inference in Regression Wei Liu (2010)
119. Robust Nonparametric Statistical Methods, Second Edition
Thomas P. Hettmansperger and Joseph W. McKean (2011)
120. Statistical Inference: The Minimum Distance Approach
Ayanendranath Basu, Hiroyuki Shioya, and Chanseok Park (2011)
121. Smoothing Splines: Methods and Applications Yuedong Wang (2011)
122. Extreme Value Methods with Applications to Finance Serguei Y. Novak (2012)
123. Dynamic Prediction in Clinical Survival Analysis Hans C. van Houwelingen and Hein Putter (2012)
124. Statistical Methods for Stochastic Differential Equations
Mathieu Kessler, Alexander Lindner, and Michael Sørensen (2012)
125. Maximum Likelihood Estimation for Sample Surveys
R. L. Chambers, D. G. Steel, Suojin Wang, and A. H. Welsh (2012)
126. Mean Field Simulation for Monte Carlo Integration Pierre Del Moral (2013)
127. Analysis of Variance for Functional Data Jin-Ting Zhang (2013)
128. Statistical Analysis of Spatial and Spatio-Temporal Point Patterns, Third Edition Peter J. Diggle (2013)
Monographs on Statistics and Applied Probability 128

Statistical Analysis
of Spatial and Spatio-
Temporal Point
Patterns
Third Edition

Peter J. Diggle
Lancaster University
England, UK
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2014 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Version Date: 20130625

International Standard Book Number-13: 978-1-4665-6024-6 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information stor-
age or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copy-
right.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that pro-
vides licenses and registration for a variety of users. For organizations that have been granted a pho-
tocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
To the memory of Julian Besag FRS, 1945-2010
This page intentionally left blank
Contents

List of Figures xv

List of Tables xxvii

Preface xxix

1 Introduction 1

1.1 Spatial point patterns . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Edge-effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Complete spatial randomness . . . . . . . . . . . . . . . . . . 9
1.5 Objectives of statistical analysis . . . . . . . . . . . . . . . . 11
1.6 The Dirichlet tessellation . . . . . . . . . . . . . . . . . . . . 12
1.7 Monte Carlo tests . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 Preliminary testing 17

2.1 Tests of complete spatial randomness . . . . . . . . . . . . . 17

2.2 Inter-event distances . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Analysis of Japanese black pine saplings . . . . . . . . 21
2.2.2 Analysis of redwood seedlings . . . . . . . . . . . . . . 21
2.2.3 Analysis of biological cells . . . . . . . . . . . . . . . . 22
2.2.4 Small distances . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Nearest neighbour distances . . . . . . . . . . . . . . . . . . 24
2.3.1 Analysis of Japanese black pine saplings . . . . . . . . 25
2.3.2 Analysis of redwood seedlings . . . . . . . . . . . . . . 25
2.3.3 Analysis of biological cells . . . . . . . . . . . . . . . . 26
2.4 Point to nearest event distances . . . . . . . . . . . . . . . . 26
2.4.1 Analysis of Japanese black pine seedlings . . . . . . . 28
2.4.2 Analysis of redwood seedlings . . . . . . . . . . . . . . 28
2.4.3 Analysis of biological cells . . . . . . . . . . . . . . . . 28
2.5 Quadrat counts . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.1 Analysis of Japanese black pine seedlings . . . . . . . 30
2.5.2 Analysis of redwood seedlings . . . . . . . . . . . . . . 31
2.5.3 Analysis of biological cells . . . . . . . . . . . . . . . . 32

ix
x Contents

2.6 Scales of pattern . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.6.1 Analysis of Lansing Woods data . . . . . . . . . . . . 33
2.6.2 Scales of dependence . . . . . . . . . . . . . . . . . . . 34
2.7 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Methods for sparsely sampled patterns 39

3.1 General remarks . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.2 Quadrat counts . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.1 Tests of CSR . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.2 Estimators of intensity . . . . . . . . . . . . . . . . . . 42
3.2.3 Analysis of Lansing Woods data . . . . . . . . . . . . 42
3.3 Distance measurements . . . . . . . . . . . . . . . . . . . . . 43
3.3.1 Distribution theory under CSR . . . . . . . . . . . . . 44
3.3.2 Tests of CSR . . . . . . . . . . . . . . . . . . . . . . . 46
3.3.3 Estimators of intensity . . . . . . . . . . . . . . . . . . 49
3.3.4 Analysis of Lansing Woods data . . . . . . . . . . . . 50
3.3.5 Catana’s wandering quarter . . . . . . . . . . . . . . . 51
3.4 Tests of independence . . . . . . . . . . . . . . . . . . . . . . 52
3.5 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . 53

4 Spatial point processes 55

4.1 Processes and summary descriptions . . . . . . . . . . . . . . 55

4.2 Second-order properties . . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Univariate processes . . . . . . . . . . . . . . . . . . . 57
4.2.2 Extension to multivariate processes . . . . . . . . . . . 60
4.3 Higher order moments and nearest neighbour distributions . 60
4.4 The homogeneous Poisson process . . . . . . . . . . . . . . . 61
4.5 Independence and random labelling . . . . . . . . . . . . . . 63
4.6 Estimation of second-order properties . . . . . . . . . . . . . 64
4.6.1 Stationary processes . . . . . . . . . . . . . . . . . . . 64
4.6.2 Estimating the pair correlation function . . . . . . . . 71
4.6.3 Intensity-reweighted stationary processes . . . . . . . 72
4.6.4 Multivariate processes . . . . . . . . . . . . . . . . . . 73
4.6.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.7 Displaced amacrine cells in the retina of a rabbit . . . . . . . 76
4.8 Estimation of nearest neighbour distributions . . . . . . . . . 78
4.8.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.9 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . 79

5 Nonparametric methods 83

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2 Estimating weighted integrals of the second-order intensity . 83
Contents xi

5.3 Nonparametric estimation of a spatially varying intensity . . 84

5.3.1 Estimating spatially varying intensities for the Lansing
Woods data . . . . . . . . . . . . . . . . . . . . . . . . 87
5.4 Analysing replicated spatial point patterns . . . . . . . . . . 90
5.4.1 Estimating the K-function from replicated data . . . . 92
5.4.2 Between-group comparisons in designed experiments . 94
5.5 Parametric or nonparametric methods? . . . . . . . . . . . . 97

6 Models 99

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2 Contagious distributions . . . . . . . . . . . . . . . . . . . . 100
6.3 Poisson cluster processes . . . . . . . . . . . . . . . . . . . . 101
6.4 Inhomogeneous Poisson processes . . . . . . . . . . . . . . . 104
6.5 Cox processes . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.6 Trans-Gaussian Cox processes . . . . . . . . . . . . . . . . . 109
6.7 Simple inhibition processes . . . . . . . . . . . . . . . . . . . 110
6.8 Markov point processes . . . . . . . . . . . . . . . . . . . . . 112
6.8.1 Pairwise interaction point processes . . . . . . . . . . 114
6.8.2 More general forms of interaction . . . . . . . . . . . . 118
6.9 Other constructions . . . . . . . . . . . . . . . . . . . . . . . 118
6.9.1 Lattice-based processes . . . . . . . . . . . . . . . . . 118
6.9.2 Thinned processes . . . . . . . . . . . . . . . . . . . . 119
6.9.3 Superpositions . . . . . . . . . . . . . . . . . . . . . . 120
6.9.4 Interactions in an inhomogeneous environment . . . . 121
6.10 Multivariate models . . . . . . . . . . . . . . . . . . . . . . . 123
6.10.1 Marked point processes . . . . . . . . . . . . . . . . . 123
6.10.2 Multivariate point processes . . . . . . . . . . . . . . . 123
6.10.3 How should multivariate models be formulated? . . . . 124
6.10.4 Cox processes . . . . . . . . . . . . . . . . . . . . . . . 125
6.10.5 Markov point processes . . . . . . . . . . . . . . . . . 128

7 Model-fitting using summary descriptions 131

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

7.2 Parameter estimation using the K-function . . . . . . . . . . 132
7.2.1 Least squares estimation . . . . . . . . . . . . . . . . . 132
7.2.2 Simulated realisations of a Poisson cluster process . . 133
7.2.3 Procedure when K(t) is unknown . . . . . . . . . . . . 134
7.3 Goodness-of-fit assessment using nearest neighbour distribu-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.4.1 Redwood seedlings . . . . . . . . . . . . . . . . . . . . 136
7.4.2 Bramble canes . . . . . . . . . . . . . . . . . . . . . . 139
7.5 Parameter estimation via goodness-of-fit testing . . . . . . . 147
xii Contents

7.5.1 Analysis of hamster tumour data . . . . . . . . . . . . 148

8 Model-fitting using likelihood-based methods 151

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

8.2 Likelihood inference for inhomogeneous Poisson processes . . 152
8.2.1 Fitting a trend surface to the Lansing Woods data . . 153
8.3 Likelihood inference for Markov point processes . . . . . . . 155
8.3.1 Maximum pseudo-likelihood estimation . . . . . . . . 156
8.3.2 Non-parametric estimation of a pairwise interaction
function . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.3.3 Fitting a pairwise interaction point process to the dis-
placed amacrine cells . . . . . . . . . . . . . . . . . . . 158
8.3.4 Monte Carlo maximum likelihood estimation . . . . . 160
8.3.5 The displaced amacrine cells re-visited . . . . . . . . . 163
8.3.6 A bivariate model for the displaced amacrine cells . . 165
8.4 Likelihood inference for Cox processes . . . . . . . . . . . . . 167
8.4.1 Predictive inference in a log-Gaussian Cox process . . 169
8.4.2 Non-parametric estimation of an intensity surface: hick-
ories in Lansing Woods . . . . . . . . . . . . . . . . . 170
8.5 Additional reading . . . . . . . . . . . . . . . . . . . . . . . . 172

9 Point process methods in spatial epidemiology 173

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

9.2 Spatial clustering . . . . . . . . . . . . . . . . . . . . . . . . 176
9.2.1 Analysis of the North Humberside childhood leukaemia
data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.2.2 Other tests of spatial clustering . . . . . . . . . . . . . 178
9.3 Spatial variation in risk . . . . . . . . . . . . . . . . . . . . . 179
9.3.1 Primary biliary cirrhosis in the North East of England 181
9.4 Point source models . . . . . . . . . . . . . . . . . . . . . . . 182
9.4.1 Childhood asthma in north Derbyshire, England . . . 185
9.4.2 Cancers in North Liverpool . . . . . . . . . . . . . . . 186
9.5 Stratification and matching . . . . . . . . . . . . . . . . . . . 189
9.5.1 Stratified case-control designs . . . . . . . . . . . . . . 189
9.5.2 Individually matched case-control designs . . . . . . . 191
9.5.3 Is stratification or matching helpful? . . . . . . . . . . 193
9.6 Disentangling heterogeneity and clustering . . . . . . . . . . 193

10 Spatio-temporal point processes 195

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

10.2 Motivating examples . . . . . . . . . . . . . . . . . . . . . . 196
10.2.1 Gastro-intestinal illness in Hampshire, UK . . . . . . . 196
Contents xiii

10.2.2 The 2001 foot-and-mouth epidemic in Cumbria, UK . 198

10.2.3 Bovine tuberculosis in Cornwall, UK . . . . . . . . . . 198
10.3 A classification of spatio-temporal point patterns and processes 199
10.4 Second-order properties . . . . . . . . . . . . . . . . . . . . . 202
10.5 Conditioning on the past . . . . . . . . . . . . . . . . . . . . 204
10.6 Empirical and mechanistic models . . . . . . . . . . . . . . . 207

11 Exploratory analysis 209

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

11.2 Animation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
11.3 Marginal and conditional summaries . . . . . . . . . . . . . . 210
11.3.1 Bovine tuberculosis in Cornwall, UK . . . . . . . . . . 210
11.4 Second-order properties . . . . . . . . . . . . . . . . . . . . . 213
11.4.1 Stationary processes . . . . . . . . . . . . . . . . . . . 213
11.4.2 Intensity-reweighted stationary processes . . . . . . . 216
11.4.3 Campylobacteriosis in Lancashire, UK . . . . . . . . . 217

12 Empirical models and methods 223

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

12.2 Poisson processes . . . . . . . . . . . . . . . . . . . . . . . . 224
12.3 Cox processes . . . . . . . . . . . . . . . . . . . . . . . . . . 224
12.3.1 Separable and non-separable models . . . . . . . . . . 225
12.4 Log-Gaussian Cox processes . . . . . . . . . . . . . . . . . . 226
12.5 Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
12.6 Gastro-intestinal illness in Hampshire, UK . . . . . . . . . . 227
12.7 Concluding remarks: point processes and geostatistics . . . . 230

13 Mechanistic models and methods 235

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

13.2 Conditional intensity and likelihood . . . . . . . . . . . . . . 235
13.3 Partial likelihood . . . . . . . . . . . . . . . . . . . . . . . . 237
13.4 The 2001 foot-and-mouth epidemic in Cumbria, UK . . . . . 238
13.5 Nesting patterns of Arctic terns . . . . . . . . . . . . . . . . 240

References 245

Index 263
This page intentionally left blank
List of Figures

1.1 Locations of 65 Japanese black pine saplings in a square of

side-length 5.7 metres (Numata, 1961). . . . . . . . . . . . . 2
1.2 Locations of 62 redwood seedlings in a square of side-length
23 metres (Strauss, 1975; Ripley, 1977). . . . . . . . . . . . . 3
1.3 Locations of 42 cell centres in a unit square (Ripley, 1977). . 4
1.4 Locations of 294 displaced amacrine cells in the retina of a
rabbit. Solid and open circles respectively identify on and off
cells. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Locations of cases of non-specific gastrointestinal symptoms
reported to NHS Direct Hampshire, UK, between 1 January
and 8 January 2001. The radius of each plotted circle codi-
fies the reporting date (smallest for 1 January, largest for 8
January). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 Realisation of CSR: 100 events in a unit square. . . . . . . . 10
1.7 The Dirichlet tessellation (——) and Delaunay triangulation
(– – –) associated with 12 points in a unit square. . . . . . . 13
1.8 Blurred critical regions for one-sided, 5% Monte Carlo tests
with s = 20, 40, 100 and s → ∞ (adapted from Marriott,
1979). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1 EDF plot of inter-event distances for Japanese black pine

saplings. —— : data; − − − : upper and lower envelopes from
99 simulations of CSR. . . . . . . . . . . . . . . . . . . . . . 20
2.2 EDF plot of inter-event distances for redwood seedlings. ——
: data; − − − : upper and lower envelopes from 99 simulations
of CSR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 EDF plot of inter-event distances for biological cells. —— :
data; − − − : upper and lower envelopes from 99 simulations
of CSR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 EDF plot of nearest neighbour distances for Japanese black
pine saplings. —— : data; − − − : upper and lower envelopes
from 99 simulations of CSR. . . . . . . . . . . . . . . . . . . 26
2.5 EDF plot of nearest neighbour distances for redwood
seedlings. —— : data; − − − : upper and lower envelopes
from 99 simulations of CSR. . . . . . . . . . . . . . . . . . . 27

xv
xvi List of Figures

2.6 EDF plot of nearest neighbour distances for biological cells.

—— : data; − − − : upper and lower envelopes from 99 sim-
ulations of CSR. . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7 Calculation of F̂ (x) for the biological cells data, using a k × k
grid of sample points for different values of k. —— : k = 7;
− − − : k = 14; ....... : k = 96. . . . . . . . . . . . . . . . . . 29
2.8 EDF plot of point to nearest event distances for Japanese
black pine saplings. —— : data; − − − : upper and lower
envelopes from 99 simulations of CSR. . . . . . . . . . . . . 30
2.9 EDF plot of point to nearest event distances for redwood
seedlings. —— : data; − − − : upper and lower envelopes
from 99 simulations of CSR. . . . . . . . . . . . . . . . . . . 31
2.10 EDF plot of point to nearest event distances for biological
cells. —— : data; − − − : upper and lower envelopes from 99
simulations of CSR. . . . . . . . . . . . . . . . . . . . . . . . 32
2.11 Locations of trees in Lansing woods: left panel, hickories; mid-
dle panel, maples; right panel, oaks. . . . . . . . . . . . . . . 34
2.12 Index of dispersion plotted against block size (k × k) for Lans-
ing Woods data, based on a 32 × 32 grid of quadrat counts.
—— : hickories; − − − : maples; ....... : oaks. . . . . . . . . 36

3.1 The nearest event, P, to an arbitrary point, O, and the nearest

event, Q, to the event P. . . . . . . . . . . . . . . . . . . . . 45
3.2 T-square sampling: the nearest event, P, to an arbitrary point,
O, and the T-square nearest neighbour, Q, to the event P. . 46
3.3 Catana’s wandering quarter: solid lines indicate a sequence of
recorded distances, starting from the point O. . . . . . . . . 52

4.1 Construction of the edge-correction weights in Ripley’s (1976)

estimator for K(t) . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Comparison between Ripley’s asymptotic approximation and
the Lotwick-Silverman formula for the sampling variance of
K̂(t). —— : n = 100; − − − : n = 200; ....... : n = 400. The
horizontal line at height 1 corresponds to equality of the two
formulae. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.3 Comparison between Chetwynd and Diggle’s estimator and
the Lotwick-Silverman formula for the sampling variance of
K̂(t). —— : n = 100; − − − : n = 200; ....... : n = 400. . . . 69
4.4 Comparison between the variances of K̃(t) and of K̂(t). ——
: n = 100; ......... : n = 200; – – – : n = 400 . . . . . . . . . . 70
4.5 Comparison between histogram-like (solid line) and kernel
smoothed estimates of the pair correlation function. Dashed
and dotted lines correspond to kernel estimates using a quartic
kernel and band-widths h = 0.038 and 0.076, respectively. . 72
List of Figures xvii

4.6 The estimate K̂(t) − πt2 for the Japanese black pine data. —
— : data; − − − : plus and minus two standard errors under
complete spatial randomness. . . . . . . . . . . . . . . . . . 74
4.7 Transformed estimates of K̂(t) for the Japanese black pine
data. —— : data; − − − : plus and minus two standard er-
rors under
√ complete spatial randomness. The left-hand panel
shows K̂(t), the right-hand panel {K̂(t) − πt2 }/t. . . . . . 75
4.8 The estimate D̂(t) = K̂(t) − πt2 for the redwood data (left-
hand panel) and for the cell data (right-hand panel). ——
: data; − − − : plus and minus two standard errors under
complete spatial randomness. . . . . . . . . . . . . . . . . . 76
4.9 Second-order properties of displaced amacrine cells. Functions
plotted are D̂(t) = K̂(t) − πt2 as follows: – – – : on cells; .......
: off cells; — — — : all cells; —— : bivariate. The parabola
−πt2 is also shown as a solid line. . . . . . . . . . . . . . . . 77
4.10 Estimates of F (x) (left-hand panel) and of G(x) (right-hand
panel) for the Japanese black pine data. —— : Ĝ(·), F̂ (·); – –
– : G̃(·), F̃ (·) . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.11 Estimates of F (x) (left-hand panel) and of G(x) (right-hand
panel) for the redwood data. —— : Ĝ(·), F̂ (·); – – – : G̃(·), F̃ (·) 81
4.12 Estimates of F (x) (left-hand panel) and of G(x) (right-hand
panel) for the cell data. —— : Ĝ(·), F̂ (·); – – – : G̃(·), F̃ (·) . 81

5.1 Estimates of the mean square error, M̂ (h) of a non-parametric

intensity estimator applied to each of the three major species
groupings in Lansing Woods. − − − − −− : hickories; − − −:
maples; ....... : oaks. . . . . . . . . . . . . . . . . . . . . . . . 87
5.2 Estimates K̂(t)−πt2 for each of the three major species group-
ings in Lansing Woods. —– : hickories; − − −: maples; .......
: oaks. Pointwise values of two standard errors assuming an
underlying homogeneous Poisson process are shown as − · − · − 88
5.3 Kernel estimates of λ(x) for the hickories (left-hand panel) and
maples (right-hand panel) in Lansing Woods, using a quartic
kernel with bandwidth h = 0.1. . . . . . . . . . . . . . . . . 89
5.4 Estimates of KI (t) for the hickories (left-hand panel) and
maples (right-hand panel) in Lansing Woods. − − − − − :
data; − − − upper and lower envelopes from 19 simulations
of the inhomogeneous Poisson process with intensity equal to
the corresponding kernel estimate λ̂(·). . . . . . . . . . . . . 91
5.5 Locations of pyramidal neurons in brain tissue sections from
12 normal subjects. . . . . . . . . . . . . . . . . . . . . . . . 92
5.6 Pooled estimate K̂(t) − πt2 from 12 control subjects, with
bootstrapped plus and minus two pointwise standard error
limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
xviii List of Figures

5.7 Locations of pyramidal neurons in brain tissue sections from

9 schizo-affective subjects. . . . . . . . . . . . . . . . . . . . 95
5.8 Locations of pyramidal neurons in brain tissue sections from
10 schizophrenic subjects. . . . . . . . . . . . . . . . . . . . 95
5.9 Pooled estimates K̂(t) − πt2 from 9 schizo-affective subjects
(left-hand panel) and from 9 schizophrenic subjects (right-
hand panel), with bootstrapped plus and minus two pointwise
standard error limits. . . . . . . . . . . . . . . . . . . . . . . 96

6.1 Realisations of two Poisson cluster processes, each with 25 par-

ents on the unit square an average of four offspring per parent
and radially symmetric Gaussian dispersion of offspring, with
parameter σ = 0.025. In the left-hand panel, each parent has
exactly four offspring. In the right-hand panel, offspring are
randomly allocated amongst the 25 parents. . . . . . . . . . 104
6.2 A realisation of an inhomogeneous Poisson process with 100
events on the unit square and λ(x) = exp(−2x1 − x2 ) . . . . 105
6.3 A realisation of a Cox process; see text for detailed explana-
tion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.4 A realisation of a log-Gaussian Cox process (dots) superim-
posed on the realisation of the underlying intensity surface
(grey-scale image); see text for detailed explanation. . . . . 110
6.5 Progressive development of a realisation of a simple sequential
inhibition process with δ = 0.08 and n = 25, 50, 75, 100 events
on the unit square. . . . . . . . . . . . . . . . . . . . . . . . 112
6.6 Realisations of two pairwise interaction point processes on the
unit square and their corresponding interaction functions (see
text for detailed explanation). . . . . . . . . . . . . . . . . . 116
6.7 Realisations of two partly inhibitory, partly attractive pair-
wise interaction point processes on the unit square and their
interaction functions (see text for detailed explanation). . . 117
6.8 A thinning of an inhibitory pairwise interaction point process
(see text for detailed explanation). . . . . . . . . . . . . . . 120
6.9 Realisations of two processes with identical second-order prop-
erties (upper panels) and their estimated nearest neighbour
distributions functions (lower panels), with dashed and solid
lines corresponding to a Poisson cluster process and the su-
perposition of a Poisson cluster process and a homogeneous
Poisson process. . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.10 An inhomogeneous pairwise interaction point process (see text
for detailed explanation). . . . . . . . . . . . . . . . . . . . . 123
6.11 A linked (left-hand panel) and a balanced (right-hand panel)
log-Gaussian Cox process (see text for detailed explanation). 127
List of Figures xix

6.12 Two mutually inhibitory bivariate pairwise interaction point

processes. The stronger inhibition is between events of oppo-
site type in the left-hand panel, and between events of the
same type in the right-hand panel (see text for detailed ex-
planation). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

7.1 The empirical sampling distribution of (log ρ̂, log σ̂) in a Pois-
son cluster process with 400 events in the unit square, ρ = 100
and σ = 0.025. . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7.2 K(t) − πt2 for the redwood seedlings. —–: data; − − −: fitted
model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.3 Goodness-of-fit of a Poisson cluster process to the redwood
seedling data, using nearest neighbour (left-hand panel) and
point-to-nearest-event (right-hand panel) distribution func-
tions. —— : data; − − − : envelope from 99 simulations of
fitted model. . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.4 A realisation of the model fitted to the redwood seedling data. 139
7.5 The empirical sampling distribution of (log ρ̂, log σ̂) for the
model fitted to the redwood seedling data. . . . . . . . . . . 140
7.6 Locations of 359 newly emergent bramble canes in a 9 metre
square plot (Hutchings, 1978) . . . . . . . . . . . . . . . . . 141
7.7 K(t)−πt2 for newly emergent bramble canes (solid black line),
fitted model (dashed black line) and 99 simulations of fitted
model cluster process (thin grey lines). In the left-hand panel,
the fitted model is a two-parameter Poisson cluster process;
in the right-hand panel, it is four-parameter thinned Poisson
cluster-process. . . . . . . . . . . . . . . . . . . . . . . . . . 142
7.8 Goodness-of-fit for the four-parameter model fitted to the
newly emergent bramble canes, using nearest neighbour (left-
hand panel) and point-to-nearest-event (right-hand panel) dis-
tribution functions. The data are shown as thick black lines,
simulations of the fitted model as thin grey lines. . . . . . . 143
7.9 A realisation of the four-parameter model fitted to the newly
emergent bramble canes . . . . . . . . . . . . . . . . . . . . 144
7.10 Locations of 359 newly emergent (solid dots) and 385 one-
year-old (open circles) bramble canes in a 9 metre square plot
(Hutchings, 1978). . . . . . . . . . . . . . . . . . . . . . . . 145
7.11 K̂ij (t) − πt2 for newly emergent and one-year-old bramble
canes: solid line corresponds to K11 (t) (newly emergent),
dashed line to K22 (t) (one-year-old), dotted line to K12 (t). . 146
xx List of Figures

7.12 Left-hand panel: K̂12 (t) − πt2 for newly emergent and one-
year-old bramble canes (thick black line) and for simulations
of fitted linked Cox process model (thin grey lines). The fit-
ted function K12 (t) − πt2 is shown as a thick dashed line.
Right–hand panel: Ĥ2 (x) for newly emergent and one-year-
old bramble canes (solid black line) and for 99 simulations of
fitted linked Cox process model (thin grey lines). . . . . . . 147
7.13 Ĥ3 (x) for newly emergent, one-year-old and two-year-old
bramble canes (solid black line) and for 99 simulations (thin
grey lines). In the left-hand panel, the simulation model is the
fitted trivariate linked Cox process model. In the centre and
right-hand panels the simulation model is random labelling. 148
7.14 Locations of 303 cell nuclei in a hamster tumour; 77 pyknotic
nuclei (solid dots); 226 metaphase nuclei (open circles) . . . 149
7.15 Dij (t) = K̂ij (t) − πt2 for the hamster tumour data, com-
pared with 99 simulations of a pair of independent homo-
geneous Poisson processes. Upper-left panel shows K̂11 (py-
knotic cells), upper-right panel K̂22 (metaphase cells), lower-
left panel K̂12 , lower-right panel K̂ (superposition). . . . . . 150
7.16 Left-hand panel: 95% confidence interval for interaction pa-
rameter θ in pairwise interaction process with h(u) = 1 −
exp{−(u/θ)2 }, fitted to the hamster tumour data. Right-
hand panel: goodness-of-fit for hamster tumour data, using
the nearest neighbour distribution function. Thick black line
is calculated from the data, thin grey lines from 99 simulations
of pairwise interaction process with θ = 0.014. . . . . . . . . 150

8.1 Fitted log-quadratic trend surfaces for the hickories (left-hand

panel) and maples (right-hand panel) in Lansing Woods . . 154
8.2 Estimates of KI (t) for the hickories (left-hand panel) and
maples (right-hand panel) in Lansing Woods. Data are shown
as thick black lines, simulations of the fitted log-quadratic in-
homogeneous Poisson process as thin grey lines. . . . . . . . 155
8.3 Non-parametric (step-function) estimate of the interaction
function for the displaced amacrine on cells. . . . . . . . . . 159
8.4 Parametric (dashed line) and non-parametric (step-function)
estimates of the interaction function for the displaced
amacrine on cells. . . . . . . . . . . . . . . . . . . . . . . . . 160
8.5 Profile log-pseudo-likelihood surface for the pairwise interac-
tion point process fitted to the on cells. The plotted value at
each point is the log-pseudo-likelihood maximised with respect
to β, holding δ and ρ fixed. . . . . . . . . . . . . . . . . . . 161
List of Figures xxi

8.6 Goodness-of-fit assessment for displaced amacrine off cells, us-

ing point-to-nearest-event (left-hand panel) and nearest neigh-
bour (right-hand panel) distribution functions. —— : data;
− − −: envelope from 99 simulations of fitted pairwise inter-
action point process. . . . . . . . . . . . . . . . . . . . . . . 162
8.7 Monte Carlo log-likelihood functions for the displaced
amacrine on cells, with interaction function (8.10), θ0 =
(δ, ρ, β) = (0.016, 0.121, 2.091) and s = 10 simulations per
log-likelihood evaluation. Thin grey lines show each of five in-
dependent replicates, thick black line the point-wise average
of the five. Left-hand panel: L̂θ0 ,10 (δ). Centre panel: L̂θ0 ,10 (ρ).
Right-hand panel: L̂θ0 ,10 (β). . . . . . . . . . . . . . . . . . . 163
8.8 Monte Carlo log-likelihood functions for the displaced
amacrine on cells, with interaction function (8.10), θ0 =
(δ, ρ, β) = (0.017, 0.122, 2.100) and s = 100 simulations per
log-likelihood evaluation. Thin grey lines show each of five in-
dependent replicates, thick black line the point-wise average of
the five. Left-hand panel: L̂θ0 ,100 (δ). Centre panel: L̂θ0 ,100 (ρ).
Right-hand panel: L̂θ0 ,100 (β). . . . . . . . . . . . . . . . . . 164
8.9 Monte Carlo log-likelihood functions for the displaced
amacrine on cells, with interaction function (8.10), θ0 =
(δ, ρ, β) = (0.017, 0.122, 2.100) and s = 1000 simulations per
log-likelihood evaluation. Thin grey lines show each of five
independent replicates, thick black line the point-wise aver-
age of the five. Left-hand panel: L̂θ0 ,1000 (δ). Centre panel:
L̂θ0 ,1000 (ρ). Right-hand panel: L̂θ0 ,1000 (β). . . . . . . . . . . 165
8.10 Schematic representation of the placement of displaced
amacrine cells in the transition from the immature to the ma-
ture retina. . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.11 Prior (smooth curves) and posterior (histograms) distribu-
tions for the parameters β, σ and φ in the LGCP model for
the hickory data. . . . . . . . . . . . . . . . . . . . . . . . . 171
8.12 Predictive probabilities P{exp[S(x)] < 1/2|data} (left-hand
panel), pointwise medians of the predictive distribution
of exp{β + S(x)} (centre panel), predictive probabilities
P{exp[S(x)] > 2|data} (right panel). Left and right panels
also show data. . . . . . . . . . . . . . . . . . . . . . . . . . 172

9.1 Residential locations at birth of 62 cases of childhood

leukaemia in North Humberside, UK (left-hand panel), and
of 143 controls sampled at random from the birth register
(right-hand panel). . . . . . . . . . . . . . . . . . . . . . . . 175
xxii List of Figures

9.2 Second-order analysis of clustering for the North Humberside

childhood leukaemia data. —— : D̂(t) for observed data; −−−
: plus and minus two standard errors under random labelling
of cases and controls. . . . . . . . . . . . . . . . . . . . . . . 177
9.3 Residential locations (postcodes) of 761 cases of primary bil-
iary cirrhosis in North East England (solid dots) and 3044
control locations selected by weighted random sampling from
the postcode directory (open circles). The dashed line in the
left-hand panel indicates the approximation to the boundary
of the study-region used in the analysis of the data. The right-
hand panel shows the densely populated part of the study-
region in more detail. . . . . . . . . . . . . . . . . . . . . . . 181
9.4 Cross-validated log-likelihood criterion for choice of band-
width in non-parametric estimation of the relative risk surface
for primary biliary cirrhosis in the North East of England. . 182
9.5 Estimated relative risk surface for primary biliary cirrhosis in
the North East of England. The scale denotes the probability,
at each location x in the study-region, that a case or control
at location x is a case. . . . . . . . . . . . . . . . . . . . . . 183
9.6 Examples of a directional model for elevation in risk around
a point source. . . . . . . . . . . . . . . . . . . . . . . . . . . 185
9.7 Study region, control locations (solid dots) and incinerator
location (solid triangle) for the North Liverpool cancer study. 188
9.8 Estimated residual spatial variation in risk for colorectal can-
cers in the North Liverpool cancer study. The risk scale is
logarithmic to base 2. Solid and dashed contours identify re-
gions within which risk is point-wise 5% significantly higher or
lower, respectively, than the average for the whole study-area. 189
9.9 Estimated residual spatial variation in risk for lung cancers in
the North Liverpool cancer study. The risk scale is logarithmic
to base 2. Solid and dashed contours identify regions within
which risk is point-wise 5% significantly higher or lower, re-
spectively, than the average for the whole study-area. . . . . 190
9.10 Estimated residual spatial variation in risk for leukaemias/
lymphomas in the North Liverpool cancer study. The risk scale
is logarithmic to base 2. Solid and dashed contours identify
regions within which risk is point-wise 5% significantly higher
or lower, respectively, than the average for the whole study-
area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

10.1 Locations (residential post-codes) of 10,572 successive cases

of non-specific gastro-intestinal disease, as reported to NHS
Direct in Hampshire, UK, between 1 January 2001 and 31
December 2003. . . . . . . . . . . . . . . . . . . . . . . . . . 196
List of Figures xxiii

10.2 Times (days since 1 January 2001) of 10,572 successive cases

of non-specific gastro-intestinal disease, as reported to NHS
Direct in Hampshire, UK, between 1 January 2001 and 31
December 2003. . . . . . . . . . . . . . . . . . . . . . . . . . 197
10.3 Data from the 2001 UK foot-and-mouth epidemic. Left panel:
small dots show the locations of all animal-holding farms in
the county of Cumbria, larger dots show the locations of those
farms that experienced foot-and-mouth during the epidemic.
Right panel: cumulative distribution of reporting times of
cases, in days since 1 February 2001. . . . . . . . . . . . . . 199
10.4 Locations of farms in Cornwall, UK, that tested positive for
bovine tuberculosis at least once during the period 1989 to
2002. The grey-scale shading of the locations corresponds to
the year in which each farm first tested positive (from white
in 1989 to black in 2002). . . . . . . . . . . . . . . . . . . . 200
10.5 Number of farms in Cornwall, UK, testing positive for bovine
tuberculosis in each year, 1989 to 2002. . . . . . . . . . . . . 201
10.6 Snap-shots from a simulated partial realisation of a continuous
spatio-temporal pairwise interaction point process with both
inhibitory and aggregated characteristics. In each frame, the
most recent 30 or 40 events are shown as solid dots, earlier
events as open circles. See text for details of the simulation
model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.7 Snap-shots from a simulated partial realisation of a continuous
spatio-temporal pairwise interaction point process with purely
inhibitory characteristics. In each frame, the most recent 30
or 40 events are shown as solid dots, earlier events as open
circles. See text for details of the simulation model. . . . . . 207

11.1 Annual numbers of incident bovine tuberculosis cases in Corn-

wall, relative to the average annual incidence over the baseline
period 1989 to 1995. . . . . . . . . . . . . . . . . . . . . . . 211
11.2 Cross-validated log-likelihood criteria, LCV (h), for kernel
smoothing over bovine tuberculosis cases in Cornwall. Thin
lines refer to comparisons between baseline period 1989 to
1995 and individual post-baseline years, whilst the thick line
is calculated as the average of the LCV (h) over all seven post-
baseline years. Dotted lines identify the location and value of
the maximised average. . . . . . . . . . . . . . . . . . . . . . 212
11.3 Kernel smoothing estimates of annual incidence of bovine tu-
berculosis cases in Cornwall, 1996 to 1999. In each case, the
comparison is between the year indicated and the seven-year
baseline period 1999 to 1995 and the common grey-scale runs
from 0 (black) to 1 (white). The value 1/7 ≈ 0.14 corresponds
to equality between current and baseline incidence. . . . . . 213
xxiv List of Figures

11.4 Kernel smoothing estimates of annual incidence of bovine tu-

berculosis cases in Cornwall, 2000 to 2002. In each case, the
comparison is between the year indicated and the seven-year
baseline period 1999 to 1995 and the common grey-scale runs
from 0 (black) to 1 (white). The value 1/7 ≈ 0.14 corresponds
to equality between current and baseline incidence. . . . . . 214
11.5 Campylobacteriosis in the district of Preston, 2000-2002. Lo-
cations of cases (left panel) cumulative distribution of report-
ing times in days since 1 January 2000 (right panel). . . . . 218
11.6 Lancashire campylobacteriosis data: study-area (left panel);
2001 population density in number of people per hectare (cen-
tre panel); locations of the 619 cases within the urban area
(right panel). . . . . . . . . . . . . . . . . . . . . . . . . . . 218
11.7 Lancashire campylobacteriosis data: kernel estimate of spatial
intensity (left panel); weekly numbers of notified cases and
fitted regression curve (right panel). . . . . . . . . . . . . . . 219
11.8 Lancashire campylobacteriosis data: K̂st (u, v) − πu2 v, ×106
(left panel); identification of sub-sets of (u, v)-space in which
spatio-temporal clustering (diagonal black hatching) and/or
spatio-temporal interaction (grey shading) is detected at the
5% level of significance (right panel). . . . . . . . . . . . . . 220

12.1 Gastro-intestinal illness in Hampshire: observed counts of re-

ported cases per day, averaged over successive weekly peri-
ods (solid dots), compared with the fitted harmonic regression
model of daily incidence (solid line). . . . . . . . . . . . . . 229
12.2 Gastro-intestinal illness in Hampshire: adaptive kernel esti-
mate of the overall spatial variation in incidence, ρ̂(x). . . . 230
12.3 Gastro-intestinal illness in Hampshire: non-parametric (solid
line) and fitted parametric (dashed line) log-transformed pair
correlation functions (left-hand panel) and temporal covari-
ance functions (right-hand panel). . . . . . . . . . . . . . . . 231
12.4 Gastro-intestinal illness in Hampshire: predictive map for 6
March 2003. The mapped value in each case is the conditional
probability that R(x, t) > 2 given the data up to and including
day t. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
12.5 Three simulated transformed Gaussian processes, R(x, t) =
ak + bk S(x, t)2k , for k = 0.5, 1, 2 (solid, dashed and dotted
lines, respectively). . . . . . . . . . . . . . . . . . . . . . . . 233

13.1 Estimated transmission kernel in the spatial SIR model fitted

to data on the 2001 FMD epidemic in Cumbria, UK. . . . . 240
13.2 Locations of Arctic tern nests (crosses) on each of two islets in
the Abro Delta National Park, Spain, and additional locations
at which height above sea-level was also recorded (triangles). 241
List of Figures xxv

13.3 Estimated interaction function, h(d), for the model fitted to

the Arctic tern nesting data (thick line) and re-estimates from
100 simulated realisations of the model (thin grey lines). . . 243
This page intentionally left blank
List of Tables

2.1 Analysis of Lansing Woods data by Mead’s procedure, using

a 32 × 32 grid of quadrat counts and 99 randomizations. A
high rank for u1 suggests aggregation. Where ties occur, the
least extreme rank is quoted. Block size refers to the number
of quadrats which are treated as a single 4 × 4 block. . . . . 35

3.1 Quadrat count analysis of Lansing Woods data, using 25

square quadrats of side 0.05. . . . . . . . . . . . . . . . . . . 43
3.2 T-square analysis of Lansing Woods data. Figures in paren-
theses indicate attained significance levels for tests of CSR
(two-sided for tN , one-sided for M ). . . . . . . . . . . . . . . 51

7.1 Estimated correlations amongst attained significance levels for

three goodness-of-fit tests . . . . . . . . . . . . . . . . . . . 136
7.2 Empirical standard errors of, and correlations amongst, pa-
rameter estimates for model fitted to newly emergent bramble
canes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

8.1 Maximised log-likelihoods for Poisson log-quadratic trend sur-

face models fitted to the Lansing Woods data . . . . . . . . 154

9.1 Deviances (twice maximised log-likelihoods) for various sub-

models fitted to the north Derbyshire childhood asthma data 186
9.2 Numbers of North Liverpool cancer cases available for analysis 187
9.3 Parameter estimates of regression effects for age, sex and
Townsend deprivation index (TI) in the North Liverpool can-
cer study, and p-values for associated likelihood ratio tests of
significance. . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

13.1 Parameter estimation for the five-parameter model fitted to

combined data from Cumbria and Devon . . . . . . . . . . . 240
13.2 Maximum partial likelihood estimates and standard errors for
the model fitted to the Arctic tern nesting data. Standard er-
rors are Monte Carlo approximations, calculated by re-fitting
the model to 100 simulated data-sets. . . . . . . . . . . . . . 243

xxvii
This page intentionally left blank
Preface

A spatial point pattern is a set of locations, irregularly distributed within

a designated region and presumed to have been generated by some form of
stochastic mechanism. In most applications, the designated region is essen-
tially planar (two-dimensional Euclidean space), but one-dimensional appli-
cations are also possible, and three-dimensional applications are becoming
more common in conjunction with the development of more sophisticated
three-dimensional scanning microscopes. The first edition of Statistical Anal-
ysis of Spatial Point Patterns appeared in 1983. Its aim was to cover the major
methodological themes of the subject and their application to data arising in
the biological sciences, especially ecology.
In the second edition, published in 2003, I extended the methodological
discussion to cover major developments in the intervening years, but also tried
to preserve the applied flavour of the book. Much of the newer work in the
area tends to be mathematically sophisticated. My aim in covering the newer
methodological developments was to discuss the central ideas without going
into the full technical detail which a rigorous treatment would require, and
which is available in the original articles. I also resisted the temptation to
discuss spatial statistics more widely. Cressie (1991) identified the three main
branches of spatial statistics as geostatistics (spatially continuous processes),
lattice processes (spatially discrete processes) and spatial point processes.
Whilst these three topics are to some extent inter-linked, they nevertheless
give rise to distinct stochastic models and associated statistical methods, and
can therefore be studied separately.
Within the realm of spatial point processes, perhaps the most important
theoretical development over the last three decades has been the provision
of formal, likelihood-based methods of inference for a reasonably wide range
of models. These have partially replaced the more ad hoc methods which
prevailed in the early 1980’s. Nevertheless, some of the ad hoc methods remain
useful, and have themselves been extended in various ways, for example in the
adaptation of non-parametric smoothing methods to spatial point processes.
New applications have also emerged and, as is usual in statistics, have in
turn motivated further methodological development. The two new areas of
application on which I draw most heavily, in my own research and in the
book, are micro anatomy and epidemiology.
In microanatomy, the points in an observed pattern typically are refer-
ence locations for cells in a microscopic tissue section. The underlying cellular
structure influences the kinds of models which are appropriate, usually in-

xxix
xxx Preface

volving the concept of interactions between near-neighbouring cells. Perhaps

more fundamentally from a statistical viewpoint, most micro-anatomical stud-
ies use a replicated sampling design in which data are obtained from several
subjects and/or several tissue sections per subject, in contrast to the tradi-
tional emphasis throughout spatial statistics on the analysis of unreplicated
patterns.
In epidemiology, the points are reference locations (typically place of res-
idence) for cases of a disease in a geographical region, often supplemented
by the reference locations for a set of controls sampled from the underlying
population at risk. The principal methodological challenge in this area of appli-
cation is to use the case-control paradigm to circumvent the obvious difficulty
of building credible parametric models for human population distributions in
a heterogeneous environment.
When the first edition was written, there were few other books available
on the then-emerging subject of spatial statistics, and none at all which dealt
exclusively with the statistical analysis of spatial point patterns. This is no
longer the case. Books on spatial point processes and associated statistical
methods include Møller and Waagepetersen (2004) and Ilian et al (2008),
whilst Van Lieshout (2000) deals exclusively with Markov point processes
and their statistical analysis. The first general text on spatial statistics was
Ripley (1981). This, and Cressie (1991), remain standard references. Other
books on spatial statistics that include substantial material on point processes
include the two volumes of Upton and Fingleton (1985, 1989), Cliff and Ord
(1981), Bailey and Gatrell (1995), Stoyan, Kendall and Mecke (1995), Waller
and Gotway (2004), Schabenberger and Gotway (2004), and Gelfand et al.
(2010). Additionally, and typically for a maturing field, a wide variety of more
specialised books have also become available. For example, Matérn (1986) is a
re-issue of Bertil Matérn’s classic 1960 Swedish doctoral dissertation that laid
many of the foundations for later developments in spatial point processes and
geostatistics. Rue and Held (2005) discuss Gaussian Markov random fields, the
most widely used modelling framework for spatially discrete processes. Chilès
and Delfiner (1999, 2012) and Diggle and Ribeiro (2007) cover the so-called
“classical” and “model-based” approaches to geostatistics, respectively.
Some topics which seemed important at the beginning of the 1980’s had
become less so by 2003, and I reduced the space given to them. One example
was the discussion of methods for sparse sampling of spatial point processes in
situ. These methods arose in the 1950’s and 1960’s, principally in connection
with field-work by ecologists, who needed to investigate the density and spatial
pattern of plant communities in the field. By 2003, these methods were rarely
used because technological advances had made much more straighforward the
task of mappping a spatial point pattern for later analysis using the more
sophisticated methods that were by then available. To my surprise, however,
these methods subsequently experienced a minor revival when they began to
be used for analysing the size and pattern of refugee encampments (see, for
example, Bostoen, Chalabi and Grais, 2007).
Preface xxxi

Thus forewarned, in this new book I have not deleted any material from
the second edition, other than to correct a number of errors, but I have added
substantial new material in places.
The biggest change from the second edition, reflected in the enlarged title,
is to discuss spatio-temporal point patterns. Spatio-temporal point process
data have long been studied in specialised fields, notably seismology (see, for
example, Zhuang, Ogata and Vere-Jones, 2002). However, in the last decade
there has been an acceleration of methodological development, accompanied
by a diversification of application as spatio-temporially indexed data have
become more widely available in many scientific fields. Book-length treatments
are now beginning to appear, including the edited collection by Finkenstadt,
Held and Isham (2007), several chapters of Gelfand et al. (2009) and, most
recently, Cressie and Wikle (2011).
Another important development, throughout the statistics discipline, has
been the rise in popularity of R as a vehicle for the dissemination of new statis-
tical methods through open-source software. Useful packages for the analysis of
spatial point process data include spatial, spatstat, MarkedPointProcess,
splancs and spatialkernel. All of these, and more, can be downloaded from
the R project web-page, www.r-project.org. I predict with some confidence
that the above list will be out-of-date by the time this appears in print.
Public-domain data-sets used in the book, and any errors of which I am
aware, can be found on the book’s web-page:
http://wwww.lancs.ac.uk/staff/diggle/pointpatternbook
My thanks are again due to many colleagues, in many places and over
some forty years, who have provided me with such stimulating working envi-
ronments, spanning the UK, Sweden, Australia and the USA. I was fortunate
to begin my career under the wise guidance of the late Prof Robin Plackett
at the University of Newcastle upon Tyne. Periods spent at the Royal College
of Forestry Stockholm, CSIRO Australia and, most recently, the University
of Liverpool, have taught me the inestimable value of working closely with
subject-matter scientists. Visits to the Department of Biostatistics at Johns
Hopkins University, Baltimore, stimulated an enduring interest in medical and
public health applications. At Lancaster University, I have been privileged to
work with a succession of talented young research students and staff, amongst
whom special mention goes to Barry Rowlingson for his patient, if doomed,
efforts over 25 years to teach me to compute efficiently.
Finally, my collaborators on the many jointly authored publications listed
amongst the references should share the credit for whatever value the book
may have, whereas responsibility for defects remains mine alone.

Peter J. Diggle, Lancaster

This page intentionally left blank
1
Introduction

CONTENTS
1.1 Spatial point patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Edge-effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Complete spatial randomness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Objectives of statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 The Dirichlet tessellation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.7 Monte Carlo tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.1 Spatial point patterns

Data in the form of a set of points, irregularly distributed within a region of
space, arise in many different contexts; examples include locations of trees in
a forest, of nests in a breeding colony of birds, or of nuclei in a microscopic
section of tissue. We call any such data-set a spatial point pattern and refer
to the locations as events, to distinguish these from arbitrary points of the
region in question.
Figures 1.1 and 1.2 show two spatial point patterns in a square region.
The first, due to Numata (1961), shows 65 Japanese black pine saplings in a
square of side 5.7 metres whilst the second, extracted by Ripley (1977) from
Strauss (1975), shows 62 redwood seedlings in a square of side 23 metres, ap-
proximately. The two patterns appear strikingly different. Figure 1.1 shows no
obvious structure and might be regarded as a “completely random” pattern, in
a sense that we shall define formally in due course. In Figure 1.2, on the other
hand, the strong clustering of seedlings requires some biological explanation
which, in this instance, is readily available. The seedlings cluster around red-
wood stumps, which are known to be present in the study region but whose
locations have not been recorded. It is important to recognise that patterns
like Figure 1.2 can arise either through some form of clustering mechanism or
through environmental variation leading to local patches with relatively high
concentrations of events. Here, as elsewhere, failure to record relevant biolog-
ical information limits the conclusions which can be drawn from a statistical

1
2 Statistical Methods for Spatial and Spatio-Temporal Point Processes

• • •• • •
• • •
• ••• • •
• • ••
• •
•• • • • •
• •• •
• • •
•• • •
•
•• • • •
•
• • • •
• • •• •
• •
• •
• • • • • •

FIGURE 1.1
Locations of 65 Japanese black pine saplings in a square of side-length 5.7
metres (Numata, 1961).

analysis. We therefore describe patterns like Figure 1.2 as “aggregated” to

avoid the mechanistic connotations of the perhaps more obvious term “clus-
tered.”
Figure 1.3 shows a further qualitatively different type of pattern, formed
in this case by the centres of 42 biological cells (Crick and Lawrence, 1975;
Ripley, 1977). The cell centres are distributed more or less regularly over the
unit square, improbably so unless there is some associated regulating mech-
anism operating to encourage an even spatial distribution of cell centres. A
possible explanation is that the cell centres are merely convenient reference
points for cells whose physical size is non-negligible relative to the scale of
observation. Quite generally, and again without wishing to imply any specific
causal mechanism, we refer to such patterns as “regular.”
The nature of the pattern generated by a biological process can be affected
by the physical scale on which the process is observed. At a sufficiently large
scale most natural environments exhibit heterogeneity, which will tend to pro-
duce aggregated patterns. At a smaller scale, environmental variation will be
Introduction 3

•
• •
• •• • • • • • •
• ••
• ••
• ••
• •
••• • •• •
••
•• ••
•
•••
••• •
• ••
•
• ••
• •••
• • •• • ••

FIGURE 1.2
Locations of 62 redwood seedlings in a square of side-length 23 metres
(Strauss, 1975; Ripley, 1977).

less pronounced and the major determinant of pattern may then be the nature
of the interactions amongst the events themselves. For example, vegetative
propagation of individual shoots will tend to produce small-scale aggregation
whereas competition for space will encourage regularity. Our classification of
patterns as regular, random or aggregated is therefore an over-simplification,
but a useful one at an early stage of analysis. At a later stage, this simplistic
approach can be abandoned in favour of a more detailed, and essentially mul-
tidimensional, description of pattern that can be obtained either by the use of
a variety of functional summary statistics or by formulating an explicit model
of the underlying process. The approach taken in this book will be to develop
methods for the analysis of spatial patterns based on stochastic models, which
assume that the events are generated by some underlying random mechanism.
Our fourth example, shown in Figure 1.4, introduces the idea of a multivari-
ate point pattern. In this example, the points represent cells of two different
types (hence, bivariate) in the retina of a rabbit. The data consist of the lo-
cations of 294 displaced amacrine cells, amongst which 152 are of a type that
4 Statistical Methods for Spatial and Spatio-Temporal Point Processes

• •
• •
•
• •
• • • •
•
• • •
• •
• • • •
• •
• •
• • •
•
• • •
• •
• •
• •
•
• • •

FIGURE 1.3
Locations of 42 cell centres in a unit square (Ripley, 1977).

transmits information to the brain when a light goes on, whilst the remaining
142 transmit information when a light goes off. The relationship between the
two component patterns can help to explain the developmental processes that
operate within the immature retina. We shall re-examine the data from this
point of view in Section 4.7.
Our fifth example is of a spatio-temporal point pattern, in which the data
provide both the location and the time of occurrence of events of scientific
interest within a specified spatial region and time-interval. Figure 1.5 shows
the residential locations and dates of 100 consecutive cases of non-specific
gastrointestinal symptoms, as reported between 1 and 8 January 2001 to NHS
Direct, a 24-hour phone-based triage service operated by the UK National
Health Service, by residents in the county of Hampshire. The cases naturally
cluster in areas of relatively high population density, but there is at least a
hint that cases close in time (circles of the same radius of nearly so) are also
closer spatially than might be expected by chance. If true, this would suggest
that multiple cases may be the result of infections from a common source.
Spatio-temporal patterns are better examined dynamically than statically.
Introduction 5

• • •
• • • • • • • • • ••
•• • • • • • • • • • • • • • •
• • • • • • • •• •
• • •• • •
• • • • • • • • •• • •
• • • •• • • • ••
• • • • • • • •• • • • •
• • • • • •• • • • • • •
• • • • • • • •• •
•
• • • • • • •• •• • • •• •
• • • • • • • •••
• • • • •• • • • •
• • • • • • •• • • • •

FIGURE 1.4
Locations of 294 displaced amacrine cells in the retina of a rabbit. Solid and
open circles respectively identify on and off cells.

The data shown in Figure 1.5 are a sub-set of a much larger data-set reported
in Diggle et al. (2003); an animation of the complete data-set by Barry Rowl-
ingson can be viewed from the book’s web-site.
We shall assume throughout this book that the spatial region of interest is
essentially planar, although most of the ideas extend, at least in principle, to
other dimensions. Even in one dimension, the distinction between temporal
and spatial point patterns is important. In the case of series of events irreg-
ularly distributed in time, for example division times in a cell proliferation
process, stochastic models and their associated statistical methods reflect the
essentially unidirectional quality of the time dimension, whereas in the corre-
sponding spatial case, for example nesting sites along the bank of a canal, no
such directionality exists. Cox and Lewis (1966) give an excellent introduction
to the analysis of temporal point patterns, whilst Daley and Vere-Jones (2002,
2005) discuss the underlying point process theory in depth.
All of our examples involve applications in the life sciences, although sim-
ilar problems arise in many other disciplines. For examples in archaeology,
astronomy and geography see, respectively, Hodder and Orton (1976), Pee-
bles (1974) and Cliff and Ord (1981). To some extent, the methods that we
6 Statistical Methods for Spatial and Spatio-Temporal Point Processes

180000
160000
140000
N−S

120000
100000
80000

400000 420000 440000 460000 480000 500000

E−W

FIGURE 1.5
Locations of cases of non-specific gastrointestinal symptoms reported to NHS
Direct Hampshire, UK, between 1 January and 8 January 2001. The radius of
each plotted circle codifies the reporting date (smallest for 1 January, largest
for 8 January).

describe remain useful (and have certainly been used) in these other areas of
application, but should not be adopted uncritically. In particular, our stochas-
tic models will be motivated by simple considerations of possible underlying
biological mechanisms that may or may not be relevant in other disciplines.

1.2 Sampling
The selection of the study region, A say, merits some discussion. In some
applications, A is objectively determined by the problem in hand, and infer-
ences are required in terms of a process defined on A itself. One example of
Introduction 7

this would be a map of all nesting sites on an island. More commonly, A is

selected from some much larger region. The selection of A may then be made
according to a probability sampling scheme, or it may simply reflect the ex-
perimenter’s view that A is in some sense representative of the larger region.
In either case, but particularly the latter, inferences drawn from an analysis
will carry much greater conviction if consistency over replicate data-sets can
be demonstrated.
As an alternative to intensive mapping within a single region A, the ex-
perimenter may choose to record limited information from a large number of
smaller regions, for example the number of events in each region. In this con-
text, the regions are called quadrats and the data are referred to as quadrat
counts. A quadrat was originally a square of side-length 1 metre, used by the
Uppsala school of plant ecologists as the basic sampling unit for investigating
plant communities in the field (Du Rietz, 1929).
Quadrat sampling remains a popular field technique in plant ecology and
elsewhere, but in some contexts it is rather impractical. This led to the de-
velopment, initially in the American forestry literature (Cottam and Curtis,
1949), of a number of distance methods for sampling spatial point patterns.
In these, the basic sampling unit is a point, and information is recorded in
the form of distances to neighbouring events, for example the distances to the
first few nearest events.
We shall refer to quadrat count and distance methods as sparse sampling
methods, to distinguish them from intensive mapping exercises. The appropri-
ate techniques for analysing data obtained by sparse sampling and by intensive
mapping are quite different. Also, analyses of sparsely sampled patterns typ-
ically have more limited objectives than do analyses of mapped patterns.
A particular form of quadrat sampling, intermediate between sparse sam-
pling and intensive mapping, consists of partitioning the study region into dis-
joint sub-regions and recording the number of events in each sub-region. Data
of this kind arise in two very different ways. The first, which has a long tradi-
tion as a method of field sampling in plant ecology, is when a rectangular study
region is partitioned into a regular grid of square or rectangular quadrats,
and a count is taken within each quadrat. The second, which typically arises
in environmental epidemiology, is when health outcome data are routinely
maintained as counts of the numbers of events in administratively defined
sub-regions. In either case, the resulting data can be represented as a realisa-
tion of a high-dimensional multivariate random variable, Y = (Y1 , ..., Yn ) say,
where Yi denotes the number of events in the ith sub-region. In this setting, a
stochastic model for the underlying point process would induce a unique sta-
tistical distribution on Y , but in practice the form of this distribution tends
to be intractable except in a few special cases. A pragmatic alternative is
to formulate a model directly for the distribution of Y , without reference to
any underlying point process. The usual method of construction is to specify
the set of conditional distributions of each Yi given all other Yj . Models of
this kind are called Markov random fields. Their construction must satisfy
8 Statistical Methods for Spatial and Spatio-Temporal Point Processes

sometimes non-obvious constraints to ensure self-consistency. Besag (1974) is

an early, and very influential, account; see also Rue and Held (2005) or the
relevant chapters of Gelfand et al. (2009) for detailed accounts.
Replicated sampling of mapped patterns is surprisingly rare. Ecological
investigations have certainly compared patterns in study regions deliberately
selected to represent different environmental conditions (see, for example,
Bagchi et al. 2011), but I am not aware of corresponding studies which have
been designed with a view to establishing the consistency of patterns in os-
tensibly similar regions. Pseudo-replication can always be achieved by par-
titioning a single region into two or more sub-regions. Genuine replication
is more common in fields such as neuroanatomy, where the natural spatial
sampling unit is a single field of view under a microscope, and there is a
well-established tradition of using hierarchical sampling designs of the form:
multiple fields of view within tissue sections; multiple tissue sections within
subjects; multiple subjects within experimental treatment groups. Studies of
this kind lend themselves to design-based inference as an alternative to the
more widely prevailing model-based inference for spatial point patterns. See,
for example, Diggle, Lange and Benes (1991), Baddeley et al. (1993) or Eglen,
Diggle and Troy (2005).

1.3 Edge-effects
Edge-effects arise in spatial point pattern analysis when, as is often the case
in practice, the region A on which the pattern is observed is part of a larger
region on which the underlying process operates. The essential difficulty is
then that unobserved events outside A may interact with observed events
within A but, precisely because the events in question are not observed, it is
difficult to take proper account of this.
For some kinds of exploratory analysis, edge-effects can safely be ignored.
We shall discuss when and why this is so at appropriate points in the text.
More generally, we can distinguish between three broad approaches to han-
dling edge-effects: the use of buffer zones; explicit adjustments to take account
of unobserved events; and, when A is rectangular, wrapping A onto a torus
by identifying opposite edges. We will illustrate each of these approaches by
considering a statistic that arises in several contexts, namely the number of
events that occur within a specified distance of an arbitrary event or location.
The buffer zone method consists of carrying out all aspects of the statistical
analysis after conditioning on the locations of all events which fall within a
buffer zone B consisting of all points less than a specified distance, d0 say, from
the edge of A. Let C = A − B denote the remainder of A after subtracting
the buffer zone. Then, it is clear that for any event or location x ∈ C, the
observed number of events within a distance d of x must equal the actual
Introduction 9

number of events in the underlying process within distance d of x, provided

d ≤ d0 , whereas for d > d0 the observed number may be less than the actual
number, thereby biasing any estimates based on these observed numbers. The
choice of d0 in the buffer zone method is awkward, since too small a value
leaves residual edge-effects, whereas too large a value effectively throws away
data unnecessarily. However, the method can be applied in adaptive form,
varying the value of d0 according to the particular statistic being used.
The adjustment method operates by making an “on average” adjustment
for the unobserved events outside A. Again using our simple example to illus-
trate, if we count the observed number, n say, of events within distance d of
a location x, and a(d) denotes the area of intersection between A and a disc
of radius d centred on the location x, then an intuitively sensible estimate
of the actual number of events within distance d of x is nπd2 /a(d). The ad-
justment method is attractive because it makes full use of the observed data,
especially when relatively large-scale effects are of interest. Note, however,
that the adjustments typically lead to an increased sampling variance for the
edge-adjusted estimator by comparison with its unadjusted counterpart. In
essence, this is an example of the common trade-off in statistical estimation
between bias and variance, as edge-corrections seek to eliminate bias at the
expense of some increase in variance.
Toroidal wrapping of a rectangular A is not so much an edge-correction
method as a trick to eliminate edge-effects in particular circumstances. Its
most common use is as a convenient way of simulating realisations of various
kinds of point process. For example, suppose that we wish to simulate a point
process model for the cell centre data shown in Figure 1.3, the most obvious
feature of which is that no two events can occur too close together. If we
attempt to simulate a process of this kind directly on a unit square A, then
points near the edge of A are favoured over points near the centre of A as
potential locations, because of the absence of potentially inhibiting events
outside A. By simulating the process on a torus and subsequently unwrapping
to a unit square for presentation, we avoid this effect. Note that it will seldom
make sense to wrap observed data onto a torus for analysis; for example, if
we were to do this with the cell centre data, then we would observe some very
small toroidal distances between pairs of cells, thus distorting the inhibitory
nature of the underlying process.

1.4 Complete spatial randomness

The hypothesis of complete spatial randomness (henceforth CSR) for a spatial
point pattern asserts that (i) the number of events in any planar region A with
area |A| follows a Poisson distribution with mean λ|A|; (ii) given n events xi
in a region A, the xi are an independent random sample from the uniform dis-
10 Statistical Methods for Spatial and Spatio-Temporal Point Processes

• • •
•• • • ••
• • • •
• •
•
•• • • ••
• •
••
• •• • • •••
• • • • •
• • •• • •• • • ••
• ••
• • •
• • •
•• • • • • •
•• • • • ••• •
•
• ••• •
• •
• • • • • •
• ••
• • • •
•

FIGURE 1.6
Realisation of CSR: 100 events in a unit square.

tribution on A. The self-consistency of (i) and (ii) is not immediately obvious,

but will be established in Chapter 4. In (i), the constant λ is the intensity, or
mean number of events per unit area. According to (i), CSR therefore implies
that the intensity of events does not vary over the plane. According to (ii),
CSR also implies that there are no interactions amongst the events. For ex-
ample, the independence assumption in (ii) would be violated if the existence
of an event at x either encouraged or inhibited the occurrence of other events
in the neighbourhood of x. In developing tests of CSR for sparsely sampled
patterns the starting point will be property (i), whilst for mapped patterns it
is more usual to start with (ii), i.e. to analyse the pattern conditional on the
observed number of events.
Intuitive ideas about what constitutes a “random pattern” can be mislead-
ing. Figure 1.6 shows a realization of 100 events independently and uniformly
distributed on the unit square. Any visual impression of aggregation is illusory.
Note also the superficial similarity to Figure 1.1.
Our interest in CSR is that it represents an idealized standard which, if
strictly unattainable in practice, may nevertheless be tenable as a convenient
Introduction 11

first approximation. Most analyses begin with a test of CSR, and there are
several good reasons for this. Firstly, a pattern for which CSR is not rejected
scarcely merits any further formal statistical analysis. Secondly, tests are used
as a means of exploring a set of data, rather than because rejection of CSR
is of intrinsic interest. Greig-Smith, in the discussion of Bartlett (1971), em-
phasized that ecologists often know CSR to be untenable but nevertheless use
tests of CSR as aids to the formulation of ecologically interesting hypotheses
concerning pattern and its genesis. Thirdly, CSR acts as a dividing hypothesis
to distinguish between patterns which are broadly classifiable as “regular” or
“aggregated”.
Another use of CSR is as a building block in the construction of more
complex models. We shall return to this topic in Chapter 6.

1.5 Objectives of statistical analysis

In any particular application, the objectives of a statistical analysis should
be determined by the scientist’s objectives in collecting the data in question.
We have already given reasons for beginning an analysis with a test of CSR.
What to do next will vary according to context.
In sparse sampling exercises, a specific objective may be to estimate the
intensity. For example, in forestry surveys an important quantity to be esti-
mated is the “stocking density,” or number of stems per hectare. The nature of
the pattern might then be of interest only in so much as it affects the sampling
distribution of the estimator.
With mapped data, the scientist will usually want a more detailed descrip-
tion of the observed pattern than can be provided by a test of CSR. One way
to achieve this is by formulating a parametric stochastic model and fitting it
to the data. If a model can be found that fits the data well, the estimated val-
ues of its parameters provide summary statistics that can be used to compare
ostensibly similar data-sets. More ambitiously, a fitted model can provide an
explanation of the underlying scientific processes. But this must involve an
element of non-statistical inference: quite apart from the obvious fact that
a model which fits the data is not necessarily correct in any absolute sense,
we shall see in Chapter 4 that a simple stochastic model for a spatial point
pattern may admit more than one scientific interpretation.
Of course, it is generally the case that modelling itself is only a means to a
wider end. A well-formulated, and well-fitting, model provides a parsimonious
description of a complex pattern, and one which will be especially useful if
its parameters can be related to scientific hypotheses about the underlying
phenomenon being studied.
Model-fitting is particularly difficult for very heterogeneous data-sets. In
such cases, it may be unhelpful to force a parametric analysis based on unten-
12 Statistical Methods for Spatial and Spatio-Temporal Point Processes

able assumptions. For example, a generic problem in environmental epidemi-

ology is to estimate the spatial variation in the risk of a particular disease,
using data on the locations of individual cases in a geographical region. One
approach to this problem might be to formulate an idealised model for the ob-
served spatial pattern of cases under the assumption that risk is spatially con-
stant, and to investigate deviations of the observed pattern from this model.
An alternative, which would be more in line with classical epidemiological
methods, would be to make a nonparametric comparison between the pattern
of cases and a second pattern of controls, defined to be a random sample from
the population at risk. We shall discuss this idea in detail in Chapter 9.

1.6 The Dirichlet tessellation

Given n distinct events xi in a planar region A, we can assign to xi a “terri-
tory” consisting of that part of A which is closer to xi than to any other xj .
This construction, referred to either as the Dirichlet tessellation or Voronoi
tessellation of the events in A, has been incorporated into stochastic models
of natural phenomena such as inter-plant competition. In these models, plants
whose territories abut are assumed to be in direct competition for available
nutrient; see, for example, Cormack (1979, pp. 171–175). For large n, the tes-
sellation is also the basis of a computationally efficient solution to a number
of problems involving the calculation of distances between events.
Except possibly along the boundary of A, each territory or cell is a convex
polygonal region. Events xi and xj whose cells share a common boundary seg-
ment are said to be contiguous. Typically, each cell vertex is common to three
cells, and the lines joining the pairs of contiguous events define a triangula-
tion of the xi , called the Delaunay triangulation. Thus, cell boundaries can be
obtained as the perpendicular bisectors of the edges of the triangulation, and
cell vertices are the corresponding circumcentres. Figure 1.7 shows a simple
example of both the tessellation and the triangulation associated with n = 12
events in a unit square. Rogers (1964) discusses the mathematical properties
of the Dirichlet tessellation in a general p-dimensional setting.
The construction of the Dirichlet tessellation and the associated Delaunay
triangulation rapidly becomes a non-trivial exercise as n increases. Green and
Sibson (1978) give a remarkably efficient algorithm whose computational cost
increases roughly as n1.5 .
Introduction 13

1.0
•
•
•
0.8
•
•
0.6

•
0.4

•
0.2

•
• • •
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIGURE 1.7
The Dirichlet tessellation (——) and Delaunay triangulation (– – –) associated
with 12 points in a unit square.

1.7 Monte Carlo tests

Even simple stochastic models for spatial point patterns lead to intractable
distribution theory, and in order to test models against data we shall make
extensive use of Monte Carlo tests (Barnard, 1963).
Quite generally, let u1 be the observed value of a statistic U and let ui : i =
2, ..., s be corresponding values generated by independent random sampling
from the distribution of U under a simple hypothesis H. Let u(j) denote the
j th largest amongst ui : i = 1, ..., s Then, under H,

P {u1 = u(j) } = s−1 , j = 1, ..., s,

and rejection of H on the basis that u1 ranks kth largest or higher gives an
exact, one-sided test of size k/s . This assumes that the values of the ui are all
different, so that the ranking of u1 is unambiguous. If U is a discrete random
variable, for example a count, tied values are possible and we then adopt the
conservative rule of choosing the least extreme rank for u1 . The extension to
two-sided tests is clear.
14 Statistical Methods for Spatial and Spatio-Temporal Point Processes

Hope (1968) gives a number of examples to show that the loss of power
resulting from a Monte Carlo implementation is slight, so that s need not
be very large. For a one-sided test at the conventional 5% level, s = 100 is
adequate.
Power loss is related to Marriott’s (1979) investigation of “blurred critical
regions”, which arise because a value of u1 which would be declared significant
in a classical test may not be declared significant in a Monte Carlo test, and
vice versa. Let the (unknown) distribution function of U under H be F (u).
For a one-sided 5% test with s = 20k, the probability that we reject H, given
that U = u1 , is


k−1
s−1

p(u1 ) = {1 − F (u1 )}r {F (u1 )}s−1−r . (1.1)
r=0
r

For a classical test, represented here by the limit s → ∞, p(u1 ) is 1 or 0

accordingly as F (u1 ) is greater or less than 0.95. The effect of the “blurring”
introduced by (1.1) is measured by Marriott’s Table 1, here reproduced as
Figure 1.8. Marriott concludes also that the extent of the blurring depends
primarily on k, so that if s = 100 is judged to be acceptable for a test at the
5% level, then s = 500 should be used for a test at the 1% level, and pro rata
for tests at smaller levels. These recommended values of s are smaller than
would be considered adequate for the estimation of P {U > u1 | H}. This is
essentially a consequence of the blurring effect. It is the rank of u1 , and not
u1 itself, which is the test statistic. A Monte Carlo test may fail to reject H
when a classical test would have done so, representing a loss of power, but
may also reject H when a classical test would have done so, representing a
gain in power.
A technical point concerning the use of Monte Carlo tests is that in practice
random sampling will be replaced by pseudo-random sampling. In the past,
we have used the generator supplied in the NAG (1977) subroutine library
or a Fortran implementation of the Wichmann and Hill (1982) generator, but
we now routinely use the built-in R function runif() (Venables and Ripley,
1994). Other references for users who wish to understand the theory under-
lying pseudo-random number generators include Kennedy and Gentle (1980)
or Ripley (1987).
A more subtle, but potentially more important, criticism is that Monte
Carlo tests encourage “data-dredging”, since the user can choose the statistic
U to focus on any seemingly aberrant feature of their data. Whilst we admit
that this is a danger, it should be obvious that “significant” results based on
pathological test statistics are of no practical value.
An inherent weakness of the Monte Carlo approach is its restriction to
simple hypotheses H. Composite hypotheses can be tested if pseudo-random
sampling is made conditional on the observed values of sufficient statistics for
any unknown parameters, but this is seldom feasible. Note that a goodness-
of-fit test that ignores the effects of estimating parameters will tend to be
Introduction 15

1.0
0.8
P(reject|u)
0.4 0.6
0.2
0.0

0.70 0.75 0.80 0.85 0.90 0.95 1.00

F (u)

FIGURE 1.8
Blurred critical regions for one-sided, 5% Monte Carlo tests with s = 20, 40,
100 and s → ∞ (adapted from Marriott, 1979).

conservative. This particular difficulty does not arise with tests of CSR for
mapped data, because the observed number of events n is sufficient for the
intensity λ, and conditional on n CSR is a simple hypothesis. But it does
affect the assessment of goodness-of-fit for more general stochastic models.
An approximate remedy, which we discuss further in Chapter 6, is to measure
goodness-of-fit by a statistic that is not directly related to the procedure used
to estimate the parameters of the model.
The principal advantage to be set against the above is that the investigator
need not be constrained by known distribution theory, but rather can and
should use informative statistics of their own choosing.
When asymptotic distribution theory is available, Monte Carlo testing
provides an exact alternative for small samples and a useful check on the ap-
plicability of the asymptotic theory. If the results of classical and Monte Carlo
tests are in substantial agreement, little or nothing has been lost; if not, the
16 Statistical Methods for Spatial and Spatio-Temporal Point Processes

explanation is usually that the classical test uses inappropriate distributional

assumptions.

1.8 Software
Spatial point pattern analysis is computationally intensive, not least because
of the heavy reliance on Monte Carlo methods of inference. As noted in the
Preface, R has become the computing environment of choice for many statisti-
cians. The splancs package (Rowlingson and Diggle, 1993) gives a wide range
of functions for statistical analysis of spatial point patterns. The Spatstat li-
brary, written by Adrian Baddeley and Rolf Turner, also implements a wide
range of methods, with a stronger emphasis than splancs on parametric mod-
elling. Many of the analyses reported in this book were implemented using a
combination of splancs, Spatstat and Voronoi (a package for computation
of the Dirichlet tessellation, written by Rolf Turner), together with some ad-
ditional functions written by the author.
More sophisticated displays than those shown in this book, for example
colour-coded overlays of point pattern maps and contour maps, can most
easily be produced using a Geographical Information System (GIS). A wide
variety of commercial and open-source GIS packages are now available. Also,
a number of R packages have been written to provide GIS-like functionality
within the R environment. For a detailed description of spatial data-handling
in R, see for example Bivand, Pebesma and Gomez-Rubio (2008).
2
Preliminary testing

CONTENTS
2.1 Tests of complete spatial randomness . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Inter-event distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Analysis of Japanese black pine saplings . . . . . . . . . . . . . . . . 21
2.2.2 Analysis of redwood seedlings . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.3 Analysis of biological cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.4 Small distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Nearest neighbour distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Analysis of Japanese black pine saplings . . . . . . . . . . . . . . . . 25
2.3.2 Analysis of redwood seedlings . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 Analysis of biological cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Point to nearest event distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.1 Analysis of Japanese black pine seedlings . . . . . . . . . . . . . . . 28
2.4.2 Analysis of redwood seedlings . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.3 Analysis of biological cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5 Quadrat counts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.1 Analysis of Japanese black pine seedlings . . . . . . . . . . . . . . . 30
2.5.2 Analysis of redwood seedlings . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.3 Analysis of biological cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.6 Scales of pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.6.1 Analysis of Lansing Woods data . . . . . . . . . . . . . . . . . . . . . . . . 33
2.6.2 Scales of dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.7 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.1 Tests of complete spatial randomness

Although complete spatial randomness is of limited scientific interest in itself,
there are several good reasons why we might begin an analysis with a test
of CSR: rejection of CSR is a minimal prerequisite to any serious attempt
to model an observed pattern; tests are used to explore a set of data and to
assist in the formulation of plausible alternatives to CSR; CSR operates as a
dividing hypothesis between regular and aggregated patterns.
In view of the above, the present discussion emphasizes two aspects: the

17
Other documents randomly have
different content
could, you know. He was a bowler if you like. I’ve bowled at Biffin for
hours an’ hours, yet if I begin to try medium the ‘work’ don’t act.”
My tutor uttered this in a tragic tone.
“I don’t care what your pace is,” said I, carried away by her beautiful
delivery, her perfect length, her “nip,” her “devil,” her break, and,
above all, her parent’s curl in the air, which was an undoubted case
of heredity, “but your bowling’s magnificent.”
“Oh, rot!” said Grace. “It ought to be faster.”
“It’s perfectly magnificent,” said I.
“Oh, rot!” said Grace again. “Do you think I don’t know when
bowling’s real A 1? Too slow for a quick-footed bat. He’s got time to
get out and hit me most horrid. Didn’t you see Archie lift me clean
over that jolly old tent. Wasn’t it a smasher? I did feel prickly. I’d kept
’em so short, and as soon as I did pitch one up a bit that’s how he
served me.”
“By Jove!” said I; “that’s what I’ll do. It’s not quite my game, you
know, but I’m hanged if I don’t go out and hit you.”
“Oh, you will,” said the enemy, with a gleam in her eye. “We’ll see
about that. Rectory rules, you know, and lots of fielders.”
Judged in this light, my new scheme was not quite so good as it at
first appeared.
“We are a pair of jays, aren’t we,” said Grace, with amazing
friendliness. “Here we’ve both gone and given ourselves away.
You’ve shown me all about your stuff, and I’ve shown you all about
mine.”
“Yes,” I said, “we’ve certainly exposed our hands. Rather a joke. But
I never thought about it at all.”
“Nor I,” said she. “But this I will say, Dimmy. Now that I’ve seen the
sort o’ tosh you do bowl, I’m certain that I shall have just a walk over.
I always guessed it was pretty bad, your bowling, but I didn’t think it
could be quite the giddy essence of utterness that it really is. If you’ll
take my tip you’ll try lobs. I might get into two minds with those, you
know, as nobody’s quite happy with lobs. Your other sort, though,
won’t have me out in a season. I should advise you to scratch. You’ll
have an awful time if you don’t. I’m speaking plain as a friend, old
chap.”
“So beastly good of you to be so beastly friendly,” said I gloomily.
The downright Grace certainly meant to behave nicely. Her advice
was perfectly well-meant and sincere; but how impossible it was to
take it! I would prefer to sacrifice my personal dignity rather than my
opportunity. Besides, her complete indifference to the result of our
encounter was a great humiliation in itself. Could she have by any
chance forgotten the stakes for which I was to play? I deemed it wise
to sound her.
“Well, I will scratch on one condition,” said I.
“My dear Dimmy,” said she, “I’m not asking a favour, you know.
Entirely in your own interests, I can assure you. You are at liberty to
play the match or scratch it as you just please. Matter of perfect
indifference to me, you know. Merely suggested scratching to spare
you a tremendous licking. Don’t matter to me personally one way or
the other, a little bit.”
“Oh, it don’t,” said I, feeling both hot and emotional; and had a
traction-engine been taking the liberty of going over me, I don’t think
I could have felt more crushed.
“Why should it?” said Grace, gazing at me with big-eyed
demureness.
“My dear Grace,” said I; “my dear Grace.”
Her eyes grew bigger.
“What’s up, old chap?” said she.
“You’re not forgetting,” I said anxiously; “you’re not forgetting, I hope,
what this match means to me. You promised to give my claims the
most serious consideration if I won, didn’t you?”
Grace’s reply was laughter. I sought to compensate my injuries a
little by persuading myself that this ebullience on the part of Grace
was in the worst possible taste. But this I knew to be a chimera,
raised from the ruins of my self-esteem; for Grace was that forthright,
fearless kind of soul who had only to do a thing to create the
precedent for it. Somehow she seemed quite unable to lose her
breeding for a moment, as, by some strange oversight, the science
of snobbery had been omitted from her education altogether.
Therefore she did not spend her time in committing the very
solecisms that she strove most to avoid. Could she have been bred
in England?
This trite reflection, Impatient Reader, is not really a digression, but
is a device introduced to allow Grace full time to have her laugh out.
When it was ended at last, I said mildly:—
“What’s the joke, Grace?”
“Why, the joke is that you’ve not got the slightest earthly, my dear
chap,” said Grace. “Else do you think I’d have taken those absurd
conditions?”
This was comforting in the extreme! It was no more than I deserved
though. But my imperial Anglo-Saxon rose in all the majesty of his
Rudyard Kipling.
“All right,” said I; “but this is going to be no walk over. I’m going all
the way, I can tell you, Grace. A game’s never won till it’s lost.”
“I’m glad you’re cheerful,” said Grace, “’cause your gruelling’ll be so
prime that you’ll want a dreadful lot of cheerfulness. It just makes me
shudder to think what’ll happen to you, Dimmy.”
“You can’t intimidate me,” said I; “you can’t make me funk you.”
“That’ll come later,” said Grace, “when you go on to bowl!”
“Cricket, cricket, and still cricket!”
It was the voice of the Rector, who had come upon us unobserved.
“Mornin’, Father,” said his daughter. “All serene this morning? You
were reading Livy rather late last night, weren’t you? Oh, you literary
men, your hours are dreadful!”
“If you are too ironical,” said the Rector, accepting the cheek she so
promptly offered him—yes, I mean in both a figurative and a literal
sense. I certainly intended no pun, but if one has to deal with the
confounded English language, how is one to avoid its pitfalls?—“If
you are too ironical,” said the Rector, “I’ll preach a fifty minutes’
sermon to-morrow, Laura.”
As was subsequently explained, this was quite the most effective
means of dealing with the misdemeanours of Miss Grace.
 CHAPTER XIX
A Case for M.C.C.
CONTRARY to expectation, breakfast was dispatched in sufficiently
reasonable time to permit my match with Grace to start about
eleven. Needless to say, Grace herself arranged the details. The
seven cricketers who were not playing on this occasion, instead of
being allowed to act the part of mere lookers-on, received orders to
field for both sides. “And, Toddles,” said Grace, in an intimidating
tone, “if you drop Dimmy and then take me, I’ll never forgive you.”
Her father, to my great uneasiness, was to be installed as scorer,
under his usual convenient willow tree; and the notorious Biffin she
proposed to nominate as umpire. Acting on the joint advice of
Toddles and Archie, I entered a formal protest against Biffin being
allowed to stand in this important fixture. This matter, which involved
much more than I had suspected, was debated at the breakfast-
table.
“Oh,” said Grace, “you’ve nothing against Biffin’s personal character,
I hope?”
“Oh, no,” said I; “it’s only that he don’t quite command my
confidence, you know.”
“How funny!” said Grace; “’cause I have every confidence in Biffin.
He knows the game, his eyesight’s good, his decisions are as
prompt as possible, and his judgment’s wonderful. Can’t expect
more of an umpire, can you? Of course, he might be better looking,
but that’s his misfortune, poor man! Besides, I never think it’s wise to
have an umpire who’s too good-looking. One’s liable to watch him,
instead of the ball, don’t you see.”
“Dimsdale, don’t you be bluffed,” said Charlie. “She’s a regular Arthur
Roberts at the game of bluff. She knows as well as anybody that
Biffin’s umpiring is worth about five hundred runs a year to her. The
darned impudent decisions I’ve seen that bounder Biffin give are
something cruel. If he’s given her in once, he’s given her in six times
when she’s been stumped yards out of her ground, simply on
account of the tip of the wicket-keeper’s nose being in front of the
wicket. Pretty barefaced to come it once, but six times is what I
should call immoral.”
“Shows his knowledge o’ the game,” said Biffin’s defender, “and his
attention to the fine points of it, too. There’s lots of umpires ’ud not
have noticed that, and I should have had to have gone out.”
“They’d not, that’s a cert.,” said Charlie; “and that you would have
had to have gone out is a dead cert.”
“I s’pose, Charlie,” said Grace, “that the tip o’ the wicket-keeper’s
nose is a part of his person, isn’t it? And Rule 42 says——”
“Here, no more cucumbers!” cried T. S. M. hastily.
“Dimsdale, don’t you budge,” said Carteret. “If you consent to Biffin,
you’ll be shamefully rooked.”
“What’s Dimmy got to do with it?” said Grace. “Is Dimmy the M.C.C.,
or what? If I say Biffin’s going to stand, Biffin will stand, and don’t you
think he won’t; ’cause if you do, you’ll be in error.”
“Go it, Lord Harris!” said Toddles. “Just hear Harris! Oh, you
autocratic person! Talk about George, why, you’re worse than a
colonel of militia!”
The case was being conducted with great fervour by both sides.
There was quite a formidable array of counsel for the prosecution.
Indeed, Grace’s defence of the indefensible Biffin had for once
caused her to stand absolutely alone. She was no whit abashed,
though. Nor did she descend to mere argument. She was thoroughly
satisfied with her own opinion, and was prepared to enforce it in the
teeth of male criticism of the most destructive kind.
“Biffin’s an unmitigated ruffian,” said Archie. “And if I can help it, he’ll
never stand again.”
“But you can’t help it, Archie,” said his sister; “’cause if I want him to
stand, he will stand, don’t you see?”
“He’s an unprincipled person,” said the little parson. “And I marvel
that Grace’s moral nature can countenance him.”
“I’ll have a bob each way on ‘moral nature,’ Toddles,” said Grace.
“His umpirin’s too thick to talk about,” said T. S. M. “Why, at Harrow
——”
“Yes, at Harrow,” said Grace. This prompt seizure of her opportunity
was of no avail, however. Public opinion was now entirely with T. S.
M. Poor Grace stood alone. She consoled herself with a massive
piece of toast, with butter and marmalade to match.
“Seeing that Dimsdale’s happiness is at stake,” said Toddles, with an
air of patronage and protection that was perfectly insufferable, “we
shall do well to stick by him in this, and give him our undivided
support. We’ll admit that he’s not much chance under the most
favourable conditions; but with Biffin as umpire he’s as good as
plucked before he goes on the field. Besides, we want this to be a
sporting event. Fair play all round, you know, and no favour, and may
the best man win.”
“Toddles,” said the keen Grace, pausing an instant in her well-
organized assault on the toast and marmalade, “you’re mixed. Sort
yourself out a bit. Toddles, you’re talking rot.”
“Oh, but, my dear Grace,” said I, “it’s not rot for me, I can assure
you. It’s a matter for earnest consideration.”
It was really enjoyable to feel such a weight of public opinion behind
one. It was evidently a crisis that had been coming slowly to a head
for years. Here was the opportunity of the long-suffering to test the
legality of Grace’s uncompromising attitude on the Biffin question. In
the somewhat technical language of the barrister, they were simply
making a test case of it, in order to get a judicial pronouncement as
to whether in future Grace was to be licensed to do as she liked.
“O’ course I shall, James,” said Grace; “I always do, don’t I?”
“I’ll tell you what,” said Archie, “I suggest that the question be
submitted to arbitration. We’ll submit it to the Guv’nor, and take his
verdict as final. Will you agree to that, Grace?”
“You know quite well, Archie,” said Grace, in a honeyed voice, “that I
am always perfectly willing for you to fill up your spare time in a way
that’s profitable and amusing to yourself, providing it’s not likely to do
you personally any harm, or to lead astray those who are younger
than you are. Talk it over by all means with your father, Archie, but if I
say Biffin will stand, you can take my word for it.”
The high-wrought state of public opinion, that was enough to make
the French pause and the Sultan tremble, merely appeared to incite
the dauntless Grace to new audacity. She positively snapped her
fingers at it, and ate her toast and marmalade with an air of the most
victorious unconcern.
“H’m,” said the little parson, in his best clerical tone; “she seems to
be a person of character and ideas. What’s to be done? We can’t let
Dimsdale be knocked down and walked over on an occasion of this
sort. Grace, I certainly think that your uncompromising attitude on
this vexed question is greatly to be deplored.”
“Deplore away,” said Grace, helping herself to butter. “What an
amusing little man you are, Toddles!”
Affairs were at a deadlock. How was it possible to negotiate if one
side would insist on having its own way?
“It’s a sort of diplomatic impasse,” said Archie. “What’s to be done?
Suppose we take Biffin by main force and put him under the
cucumber frame, and keep him fastened down? There’d be no more
bad decisions then.”
“Plenty of bad language, though,” said some person of wit.
“And we wouldn’t release him,” said Carteret, “until Grace had
actually played an innings without damaging the eyesight of the
cucumbers or otherwise mutilating them.”
“I call Grace’s behaviour beas’ly bad form,” said the Harrow captain.
“Oh, I know it’s beastly bad form,” said Grace; “but then, you see,
Tommy, one has to play pretty low down to gain the appreciation of
one’s family.”
Had Grace’s cause only been a just one, the manner in which she
maintained it against all comers must have evoked unqualified
admiration. The cabal was powerless in the face of her despotic
attitude. They said hard things, and they said rude things, as
brothers will, even if they have a sister who is a first-class angel of
an unimpeachable appearance. But although Grace stood alone,
discredited, out of favour, a fallen idol, and a mark for some very
cutting observations, mostly the Harrow captain’s, who saw his
moment, and, boylike, was exulting in it—despite all this, Grace
continued to consume toast and marmalade as valiantly as ever.
Now and then she diversified this proceeding by looking daggers at
Toddles, when that irrepressible little clergyman made faces at her.
Now and then she introduced a brief remark on her own account,
that on examination proved to be as flinty and hard-edged as a chip
of granite. It was plain that the exercise of a considerable force of
character had been the secret of Grace’s ascendancy and pre-
eminence in regard to these great men. And having obtained her
power, she did not hesitate to abuse it, as they say her sex generally
do.
“In my opinion,” said Archie, “there’s been a mistake in Grace’s
destiny. Her arrogance and sweet unreasonableness makes her look
a bit out of drawing, I think. Strikes me that Nature planned her for a
Gladstone or a Mailyphist, and then made her a girl for fun. But I
believe she simply doesn’t care what we think of her.”
“Oh, yes, I do, you know, Archie,” said Grace; “I’m as cut up as can
be. I’m quite put off my game. You had better let us have some more
toast, Jane. Toddles, pass the marmalade—and the butter. Yes, I
think I’ll have the butter, too.”
However, in my eyes Grace’s splendid isolation had such a nobility,
such a dignity, such a pathos of its own, that it struck me with some
suddenness that a little magnanimity might not be altogether out of
place. It was patent, however, that her brothers had such firm
convictions on the point at issue that they were not likely to exercise
it. Therefore, I had a try myself.
“My dear Grace,” said I, “don’t let’s worry about Biffin any more. I’m
perfectly willing for him to stand, you know.”
“What!” cried the whole table with one voice.
Yet I ask you what could a fellow do under the circumstances?
Splendid isolation is magnificent, of course, but not being one of
Grace’s brothers, how could I help pitying the isolated?
The storm of contumely that my unconditional surrender provoked
was woeful. Even the gallant Optimist reviled me. Their unanimity
was crushing. It was not the question at issue that mattered so
much; it was the general principle. It always is the general principle.
They considered themselves betrayed. They had pledged
themselves and their interests entirely in my cause, and then I calmly
go over to the other side and merge that cause in the enemy’s. In
fact, in the impassioned language of Toddles, the more they
examined the fine points of my conduct, the deeper the iron entered
into their souls.
“Jolly good o’ the iron,” said Grace; “improve ’em no end. Been
wanting a tonic a long time.”
Grace, indeed, I am glad to state, took an entirely different view of
my behaviour. Never had I seen her face so brightly eloquent as
when she laid down her coffee-cup and looked at me.
“Dimmy,” said she, “you’re a good sort, that’s what you are—a
ripping good sort. Dimmy, you’re A 1!”
Her tone implied that she meant it, too. And really it was decidedly
consoling to feel that we stood together facing scorn and disfavour
on every hand. But it seemed that the tactful Grace knew how, when,
and where to be generous. Or, no, I’m quite sure her generosity was
not studied at all. It was just unaided nature!
“Look here, you men,” said she; “as Dimmy’s such a good sort, he’s
not going to be such a good sort, do you see? No, I don’t mean that
exactly; I mean——”
“If it’s sheer cheek that you mean, which we’ve every reason to fear
that it is,” said Toddles, “we shall be very grateful if you’ll be content
to consider it said. If we have any more before lunch I’m thinking that
some of the batting won’t be of a very high order at Tonbridge on
Monday.”
“That’ll do, Toddles,” said Grace. “Have a rest now, there’s a good
little man. What was I saying? Oh, as Dimmy’s done the right thing,
I’m going to do the right thing, too. Father, will you stand to-day,
please?”
A rousing cheer greeted this announcement.
“I’ve always said,” remarked George, “that more can be done by the
kindness method in the treatment of these wild natures than cruelty,
firearms, and that sort of thing. Here’s Grace lying down now,
apparently as tame and docile as a kitten, without the use of red-hot
irons or anything of the kind. And it’s so much better than burning
her fur, don’t you know.”
The Rector consenting to stand, the affair terminated. Biffin, for that
day at least, had to be content with the humbler functions of the
scorer.
Breakfast over, we trooped out into the sun. And as we did so I am
free to confess that my attire was a trifle irregular. Carteret being the
most medium-built man amongst us, except in the matter of girth,
and, therefore, the most resembling me, had very kindly lent me his
buckskins; Charlie lent me one of his shirts, which, to my infinite
pleasure, he assured me was the one he wore in the ’Varsity match,
when he got so many Oxford wickets, and paid so very little for them;
while Archie subscribed the identical pair of unmentionables in which
he made his record score against Sussex at Brighton last year. In
passing, it should be noted that record scores have quite a habit of
getting themselves made against Sussex at Brighton. It is probably
the sea air.
Despite this peculiar and rather extensive outfit, and an unbiassed
umpire withal, I was earnestly assured that I could not possibly have
a look in. In fact, Grace was popularly supposed to be invincible in
single combat under Rectory rules, on the Rectory wicket. A copy of
the rules aforesaid was duly deposited in the Rector’s hands before
the match began. And, although I was privileged to peruse them, the
one conclusion to which they helped me was, that although the laws
of the M.C.C. had very kindly and thoughtfully provided nine several
ways in which a batsman might get out, those of the Rectory had
most generously furnished nine and ninety.
I’ll admit at once that I had not the least confidence in myself.
Everybody took a simple-minded pleasure in telling me that Grace
never had been beaten single-handed on her own playing-piece.
“My boy,” said the little parson, with his excruciating friendliness,
“that brown-haired, brown-faced, brown-booted, brown-hollanded
person, familiarly known as Grace, is as full of wiles, tricks, and low
devices as a certain person with a toasting-fork and a curly tail whom
I shall not even permit myself to mention. If you can take Grace
down a peg, posterity becomes your servant. Your fame will be for all
time, your name will be in Wisden. For I believe I’m right in saying
that under these conditions Grace would knock spots off half the
Middlesex eleven.”
“Certain of it,” said Archie pleasantly. “She’s a holy terror; and the
charm of it is, Dimsdale, that she never has made a secret of what
she’s going to do to you.”
“Why should I?” said Grace. “I’m going to give him a most awful
licking for last night’s horrible behaviour.”
“All right,” said I meekly; “lick away.”
“Dimmy,” said she, “if I don’t, I’ll give you a scarf-pin.”
“I shall require more than a scarf-pin,” said I. And the emphasis I
used was unmistakable.
“By Jove!” said Grace, “I was forgetting that. Thanks for the
reminder, old chap. Will you call, or shall I?”
Next moment Grace had won the toss.
“Dimmy,” said she, “as your bowling’s so thoroughly depressing, and
you’re not likely to have me out in a year, I think you’d better bat first.
Get your pads on.”
 CHAPTER XX
A Case for Another Eminent Authority
WHEN I buckled on Toddles’ pads I had all the symptoms of a bad
attack. I was a trifle dizzy, I was half blind, my heart just seemed to
be trying what it could do, while my limbs were equally irresponsible.
There was a great jug of cider laid on a table under the trees, and
the Optimist would insist on my taking a draught of it, ere I went in to
bat.
Grace placed her field with consummate care. Everybody was
laughing as I took my guard.
“Don’t think he’ll stay three minutes,” said Carteret, quite audibly.
Had Carteret only known, it was the kind of remark to make me stay
three hours. That slice of Anglo-Saxon in my constitution, that I have
already had occasion to advertise, objects above all things to be
walked over rough-shod. I knit my teeth. I determined to perish or
prevail. That moment, though, I should have been very well satisfied
had the perishing anticipated itself by occurring there and then.
Grace’s bowling was much as I had judged it to be. I knew her
hereditary peculiarities would take a terrible amount of negotiating.
And they certainly did. The very first ball I turned half round to, with
the idea of getting it away to leg, whereon the flight, slow as it was,
so deceived me that had it not been for my exceedingly thoughtful
and well-trained right leg I should have suffered the humiliation of
being clean bowled middle stump.
“How’s that?” cried Grace.
“Didn’t pitch straight,” said the Rector.
I sighed my deep relief.
“There she goes,” said T. S. M. from extra cover. “Begins her games
at once. If Biffin had been standin’ it would have been ‘Hout, Miss!’
sure as a gun. Lucky for you, Dimsdale.”
“Tommy,” said Grace, “will you have the goodness to change places
with Toddles at short-leg? Very close in please, Tommy. I’m going to
bowl a few half-volleys just outside the leg stick; so you will look out
for your face, won’t you? And you won’t funk ’em, will you, Tommy?
And young boys shouldn’t be quite so jolly cheeky, should they?”
In addition to her curl, there were several other things appertaining to
Grace’s bowling that required watching. Her length was perfect, and,
strangely enough, like her model the great Alfred Shaw, she had
acquired the trick of heightening and lowering her delivery without
any appreciable change in the action, but a pretty considerable
amount in the flight. And, better than this, or worse, she was
mistress in a measure of the painfully difficult art of making
occasional balls “hang.” Although the Rectory wicket was well-nigh
perfect, one had to watch her all the way, and then be prepared to
alter one’s tactics at the last moment. She could make them “do a
bit” both ways, and, in addition to all these accomplishments, she
had the imperturbable temperament of the really great bowler—she
didn’t mind being hit. That attitude of mind is undoubtedly the hall-
mark of the master. She kept pitching up to me in the most
audacious way. But I resolutely refused to “have a go,” until at last
she had the downright impudence to send me a particularly slow full
toss to leg, which I, of course, promptly cracked to the fence for four.
“Thought you wouldn’t be able to resist that,” she said winningly.
“And do have a smack at this, Dimmy, just for fun.”
“This” was a particularly silly-looking half-volley well on the leg-side
also. Having tasted the delights of a fourer so recently, I was
naturally a bit headstrong and uplifted. I had a full sweep at it, and in
the heat of the moment utterly ignored the fatal curl. As a
consequence I caught it on the extreme end of my bat, and it went
spinning up a considerable height, straight into the hands of mid-on.
My very soul groaned. To be caught napping so absurdly and so
palpably! My emotions were so bitter that gall becomes honey by
comparison. For I had walked into the trap with my eyes open.
Now the Optimist was the fieldsman at mid-on. And the dear, kind
Optimist, most unselfish of men, had a fellow feeling that made him
wondrous kind! The Optimist shaped for the catch in the crudest
manner. He dropped me inexcusably in consequence. It was idle of
him to urge, as urge he did, that the sun was in his eyes, and that he
couldn’t see the catch. As the bowler fiercely pointed out, the sun
was directly behind him.
“It must have been the shadow, then,” said the Optimist unblushingly.
The roars of laughter that greeted his unscrupulous behaviour and
his subsequent effrontery were infectious. Even the Rector
contributed a hearty guffaw.
“Little Clumpton’s sold you this time!” cried T. S. M. in ecstasy. “You
may be very clever, Grace, but you’ve just got left.”
The bowler’s dignity and self-restraint were really very fine, however.
“He’ll simply get it all the worse when I go in,” was her Spartan
answer.
“We shall all take extremely great care to collar anything you put up,
though,” said T.S.M., “so you’d better play piano till you’ve got the
runs off.”
Grace continued to bowl even wilier and slower than before. Runs
were very difficult to obtain, but, nevertheless, I warily, cautiously
obtained a few. The bulk of them were made by means of leg
touches and pushes, and occasional big singles into the country.
She was too slow to cut; behind-the-wicket strokes were, by Rectory
rules and the laws of single wicket also, ineligible. But I was able
once to regale myself with a hit past cover for three. This was the
only time, however, that I got a chance to play my favourite stroke,
as the bowler was evidently of opinion that it was too expensive to
feed. I had made twenty-three by careful play when I got into two
minds with one that curled outrageously, and hung as well. I returned
it as tamely as possible to the bowler, who clasped it lovingly and
said: “Poor old Dimmy! Did ’um, then!”
Thereon she walked off under the trees to a little light and liquid
refreshment, which for her partook of the nature of that innocuous
concoction known as stone-ginger; whilst I ruefully unbuckled the
pads of the ironical Toddles.
All things considered, I felt that I had no reason to be dissatisfied
with my score. Twenty-three was quite the maximum of what I had
expected to get, as from the first I had not been disposed to under-
rate the excellence of Grace’s bowling. Indeed, she was kind enough
to say herself, in a reflective tone,—
“Your batting’s really very decent, Dimmy, very decent, indeed, you
know. So glad you watch the ball. Strikes me you’re the sort o’ man
to get runs on a bad wicket. With a bit more experience you ought to
do things. Oughtn’t he, boys?”
“Oh, of course,” said T. S. M. “If he can get twenty-three against your
bowlin’ he must be phenomenal. Reg’lar freak—fit for Barnum!”
“You’ve never got twenty-three against it, Tommy, anyhow,” said
Grace.
“Such a beas’ly bore, don’t-cher-know,” said the Harrow captain
wearily, “to keep hittin’ girls to the fence, and then havin’ to go and
fag after it for ’em.”
“Why, you know quite well, Tommy,” said his sister, in a very pained
voice, “that I’ve never let you fetch a ball for me in your life. No,
never. It’s shameful of you, Tommy, to talk like that, ’cause it’s not
true. Look here, you men, it’s not true, is it?”
“Oh, dear, no,” said Toddles. “You’d scorn to do it, Grace. We all
know that. You’re far too good a sportsman—I mean sportswoman.”
“Stick to the man,” said Grace. “Sounds so much primer, somehow.”
“What’s Toddles up to now?” said Charlie suspiciously. “Whenever
he talks to Grace in that kind of way there’s something behind it.
Does he want to smoke in the drawing-room, or is it breakfast in
bed? Grace, distrust that man. Last time he was allowed to put sugar
in his tea with his fingers, instead of the tongs.”
“I like that,” said Grace. “Why, that’s what you all do, you horrible
creatures! Even Dimmy does.”
“‘Even Dimmy does!’” repeated Archie. “That’s your batting, my boy.”
And as I actually saw Grace blush at Archie’s pointed remark I
began to persuade myself that it really must be my batting.
When Grace went in, she did not put on pads, for a sufficient reason,
but it amused us all, and particularly her parent, to see her don a
right-hand batting glove.
“It’s all right, father,” said she. “Sha’n’t need it, of course. It’s only out
of respect for Dimmy, you know. Looks a bit cheeky to go in with
nothing on, as though you were only playing golf or marbles, or
something like that.”
“Or having a bath,” said Toddles, sotto voce.
It was characteristic of Grace that she never held people guilty of
laughing directly at her. And I am not sure, either, whether this
simpleness of mind did not spring from a sublime faith in herself and
all her works. Certainly when she set about getting the twenty-four
runs necessary for my defeat, she proceeded to wipe them off in a
magnificently confident manner. My first three balls yielded five.
This certainly would not do. I must try lobs. But why, oh, why had my
youth been so grievously misspent? Oh, why, I asked myself in the
bitterness of my spirit had I always been bat in hand at the nets,
slogging away for hours, instead of doing now and then a little
honest bowling? It made me giddy to think of what service a decent
length and a fair command of the ball would be to me at this
moment. Oh, if I could only bowl! If I could only bowl! Young men, I
exhort you to heed these awful consequences. Batting in itself is
very alluring, but there are other things in cricket besides a cut for
four, delightful as that is. When the other side are in, it is well to have
a dim idea of how to get them out. At this dread hour, owing to the
errors of my childhood, I had not, alas! the remotest notion how to do
so.
Nevertheless, the veiled jeers of the field, the frank amusement of
the umpire, and the downright contempt of the person wielding the
willow, made my Anglo-Saxon once more rise within me. Grace’s
does-he-call-this-bowling air was most exasperating. But I still went
on in my dogged, defiant, get-there-sometime style. I might be
without hope, but I was determined that the enemy should not know
it.
Bowling slow, elementary, underhand twisters, I kept running after
them up the pitch in a frankly dare-devil manner, and several times
took red-hot cracks travelling to mid-off about ten yards from Grace’s
bat. Runs continued to come, however, just as they thought fit
apparently, but my fielding was so whole-hearted that broad grins
presently succeeded derisive smiles on the faces of those who
witnessed it. But the five became fifteen in no time. Nine more and
all was over. The imminence of disaster nerved me to superhuman
efforts. Grace mistimed one ball a little, and as it rose from her bat
for a short distance, I sprawled arms and legs up the pitch, and
literally hurled myself at it. I just contrived to touch it with my finger-
tips as it fell. Had it come off it would have been something to talk
about; as it was, it cost divers seasoned cricketers a blink of
astonishment.
 “I just contrived to touch it.”
Willow, the King.] [Page 306.

“Dimmy,” said the one wielding the willow, “aren’t you afraid o’ your
backbone at all?”
It was apparent that she was becoming impressed. With that thought
I recalled the words of the penetrating Archie: “All women have their
weakness. Grace’s is for good fielding.” I must show her what I could
do. After vainly striving to reach one that she pulled well wide of mid-
on for three, she said,—
“Dimmy, please don’t do that. It worries me. I’m so afraid that you’ll
twist yourself into something that could never be untied. That would
be horrible, wouldn’t it? And I’m so afraid of your backbone.”
“It is in your own power,” said I, “to end these gyrations. You have
only got to get out, you know.”
Her score had now reached twenty. Four more and a life’s happiness
was wrecked. Hope there was none. She was thoroughly set, and
capable of doing anything with the miserable stuff that I was rolling
up. It was in vain that I altered the position of the field after the
delivery of every ball. She inevitably dispatched the one that followed
past the precise place from which the man had just been taken. Her
batting was really cruel in its complacence and resource. The grim
gleam that illuminated her look knocked at my heart. A gleam does
not knock as a rule, I know, but many and strange things are allowed
to happen to the heart of a man in my desperate predicament. The
light-minded fieldsmen thought it quite a joke, however, and they
proferred no end of wise suggestions. Had I not better have my point
a bit squarer; my mid-off a bit deeper; my extra cover a bit more
round; and the two men guarding the cucumber frame standing in
front of the cucumbers, instead of sitting on the woodwork?
I thanked them in a chastened tone for being so very helpful.
My pitiful bowling has already been the subject of various painful
home truths. But I do believe that my fielding was not unworthy of
kind phrases. At least, it argues unusual excellence to gain the open
approbation of the great. Yet when I stopped three smashing half-