(FREE PDF Sample) Statistical Inference and Simulation For Spatial Point Processes 1st Edition Jesper Møller Ebooks

Full download ebook at ebookgate.
com
Statistical inference and simulation for

spatial point processes 1st Edition Jesper
Møller
https://ebookgate.com/product/statistical-
inference-and-simulation-for-spatial-point-
processes-1st-edition-jesper-moller/
Download more ebook from https://ebookgate.com

More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Statistical Inference for Fractional Diffusion

Processes Wiley Series in Probability and Statistics
1st Edition B. L. S. Prakasa Rao
https://ebookgate.com/product/statistical-inference-for-
fractional-diffusion-processes-wiley-series-in-probability-and-
statistics-1st-edition-b-l-s-prakasa-rao/
Statistical Inference 2nd Edition George Casella
https://ebookgate.com/product/statistical-inference-2nd-edition-
george-casella/
Models for Probability and Statistical Inference Theory

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

and Applications 1st Edition James H. Stapleton
https://ebookgate.com/product/models-for-probability-and-
statistical-inference-theory-and-applications-1st-edition-james-
h-stapleton/
Principles of Statistical Inference D. R. Cox
https://ebookgate.com/product/principles-of-statistical-
inference-d-r-cox/
Essentials Of Statistical Inference 1st Edition G. A.

Young
https://ebookgate.com/product/essentials-of-statistical-
inference-1st-edition-g-a-young/
Principles of Statistical Inference 1st Edition D. R.

Cox
https://ebookgate.com/product/principles-of-statistical-
inference-1st-edition-d-r-cox/
Statistical Inference A Short Course 1st Edition

Michael J. Panik
https://ebookgate.com/product/statistical-inference-a-short-
course-1st-edition-michael-j-panik/
Instructor s Solutions Manual for Probability and

Statistical Inference 7e 7th/Rev e. Edition Robert Hogg
https://ebookgate.com/product/instructor-s-solutions-manual-for-
probability-and-statistical-inference-7e-7th-rev-e-edition-
robert-hogg/
CAT#C2654_TitlePage 7/2/03 9:41 AM Page 1
Statistical Inference
and Simulation
for Spatial Point
Processes
Jesper Møller
Rasmus Plenge Waagepetersen
CHAPMAN & HALL/CRC

A CRC Press Company
Boca Raton London New York Washington, D.C.
© 2004 by Chapman & Hall/CRC

C2654 disclaimer Page 1 Monday, August 18, 2003 2:02 PM
Library of Congress Cataloging-in-Publication Data
Møller, Jesper.
Statistical inference and simulation for spatial point processes / Jesper Møller and
Rasmus Plenge Waagepetersen.
p. cm.
Includes bibliographical references and index.
ISBN 1-58488-265-4 (alk. paper)
1. Point processes. 2. Spatial analysis (Statistics) I. Waagepetersen, Rasmus Plenge. II.
Title.
QA274.42.M65 2003
519.2¢3—dc22 2003058463
This book contains information obtained from authentic and highly regarded sources. Reprinted material
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
efforts have been made to publish reliable data and information, but the author and the publisher cannot
assume responsibility for the validity of all materials or for the consequences of their use.
Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, microÞlming, and recording, or by any information storage or
retrieval system, without prior permission in writing from the publisher.
The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for
creating new works, or for resale. SpeciÞc permission must be obtained in writing from CRC Press LLC
for such copying.
Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identiÞcation and explanation, without intent to infringe.
Visit the CRC Press Web site at www.crcpress.com
No claim to original U.S. Government works

International Standard Book Number 1-58488-265-4
Library of Congress Card Number 2003058463
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
Contents
Preface
Acknowledgments
1 Examples of spatial p oint patterns
2 Intro duction to p oint pro cesses

2.1 Point processes on Rd
2.2 Marked point processes and multivariate point processes
2.3 Unified framework
2.3.1 Characterisation using void events
2.3.2 Characterisation using the generating functional
2.3.3 The standard proof
2.4 Space-time processes
3 Poisson p oint pro cesses

3.1 Basic properties
3.1.1 Definitions
3.1.2 Existence and independent scattering property
3.1.3 Constructions of stationary Poisson processes
3.2 Further results
3.2.1 Slivnyak-Mecke’s theorem
3.2.2 Superpositioning and thinning
3.2.3 Simulation of Poisson processes
3.2.4 Densities for Poisson processes
3.3 Marked Poisson processes
3.3.1 Random independent displacements of the points
in a Poisson process
3.3.2 Multivariate Poisson processes and random la-
belling
4 Summary statistics
4.1 First and second order properties
4.1.1 Basic definitions and results

4.1.2 The second order reduced moment measure
4.2 Summary statistics
4.2.1 Second order summary statistics
4.2.2 Directional K-functions
4.2.3 Summary statistics based on interpoint distances
4.3 Nonparametric estimation
4.3.1 Nonparametric estimation of intensity functions
4.3.2 Nonparametric estimation of K and L
4.3.3 Edge correction
4.3.4 Envelopes for summary statistics
4.3.5 Nonparametric estimation of g
4.3.6 Nonparametric estimation of F , G, and J-functions
4.4 Summary statistics for multivariate point processes
4.4.1 Definitions and properties
4.4.2 The stationary case
4.4.3 Nonparametric estimation
4.5 Summary statistics for marked point processes
5 Cox pro cesses

5.1 Definition and simple examples
5.3 Neyman-Scott processes as Cox processes
5.4 Shot noise Cox processes
5.4.1 Shot noise Cox processes as cluster processes
5.4.2 Relation to marked point processes
5.4.3 Examples
5.4.4 Summary statistics
5.5 Approximate simulation of SNCPs
5.6 Log Gaussian Cox processes
5.6.1 Conditions on the covariance function
5.7 Simulation of Gaussian fields and LGCPs
5.8 Multivariate Cox processes
5.8.2 Multivariate log Gaussian Cox processes
5.8.3 Multivariate shot noise Cox processes
6 Markov p oint pro cesses

6.1 Finite point processes with a density
6.1.1 Papangelou conditional intensity and stability
conditions
6.2 Pairwise interaction point processes
6.2.1 Definitions and properties

6.2.2 Examples of pairwise interaction point processes
6.3 Markov point processes
6.3.1 Definition and characterisation
6.3.2 Examples
6.3.3 A spatial Markov property
6.4 Extensions of Markov point processes to Rd
6.4.1 Infinite Gibbs point processes
6.5 Inhomogeneous Markov point processes
6.5.1 First order inhomogeneity
6.5.2 Thinning of homogeneous Markov point processes
6.5.3 Transformation of homogeneous Markov point
processes
6.6 Marked and multivariate Markov point processes
6.6.1 Finite marked and multivariate point processes
with a density
6.6.2 Definition and characterisation of marked and
multivariate Markov point processes
6.6.3 Examples of marked and multivariate Markov
point processes
6.6.4 Summary statistics for multivariate Markov point
processes
7 Metrop olis-Hastings algorithms

7.1 Description of algorithms
7.1.1 Metropolis-Hastings algorithms for the conditional
case of point processes with a density
7.1.2 Metropolis-Hastings algorithms for the uncondi-
tional case
7.1.3 Simulation of marked and multivariate point
processes with a density
7.2 Background material for Markov chains
7.2.1 Irreducibility and Harris recurrence
7.2.2 Aperiodicity and ergodicity
7.2.3 Geometric and uniform ergodicity
7.3 Convergence properties of algorithms
7.3.1 The conditional case
7.3.2 The unconditional case
7.3.3 The case of marked and multivariate point pro-
cesses
8 Simulation-based inference
8.1 Monte Carlo methods and output analysis

8.1.1 Ergodic averages
8.1.2 Assessment of convergence
8.1.3 Estimation of correlations and asymptotic vari-
ances
8.1.4 Subsampling
8.2 Estimation of ratios of normalising constants
8.2.1 Setting and assumptions
8.2.2 Exponential family models
8.2.3 Importance sampling
8.2.4 Bridge sampling and related methods
8.2.5 Path sampling
8.3 Approximate likelihood inference using MCMC
8.3.1 Some basic ingredients in likelihood inference
8.3.2 Estimation and maximisation of log likelihood
functions
8.4 Monte Carlo error
8.5 Distribution of estimates and hypothesis tests
8.6 Approximate missing data likelihoods
8.6.1 Importance, bridge, and path sampling for missing
data likelihoods
8.6.2 Derivatives and approximate maximum likelihood
estimates
8.6.3 Monte Carlo EM algorithm
9 Inference for Markov p oint pro cesses

9.1 Maximum likelihood inference
9.1.1 Likelihood functions for Markov point processes
9.1.2 Conditioning on the number of points
9.1.3 Asymptotic properties of maximum likelihood
estimates
9.1.4 Monte Carlo maximum likelihood
9.1.5 Examples
9.2 Pseudo likelihood
9.2.1 Pseudo likelihood functions
9.2.2 Practical implementation of pseudo likelihood
estimation
9.2.3 Consistency and asymptotic normality of pseudo
likelihood estimates
9.2.4 Relation to Takacs-Fiksel estimation
9.2.5 Time-space processes
9.3 Bayesian inference
10 Inference for Cox pro cesses

10.1 Minimum contrast estimation
10.2 Conditional simulation and prediction
10.2.1 Conditional simulation for Neyman-Scott processes
10.2.2 Conditional simulation for SNCPs
10.2.3 Conditional simulation for LGCPs
10.3 Maximum likelihood inference
10.3.1 Likelihood inference for a Thomas process
10.3.2 Likelihood inference for a Poisson-gamma process
10.3.3 Likelihood inference for LGCPs
10.4 Bayesian inference
10.4.1 Bayesian inference for cluster processes
11 Birth-death pro cesses and p erfect simulation

11.1 Spatial birth-death processes
11.1.1 General definition and description of spatial birth-
death processes
11.1.2 General algorithms
11.1.3 Simulation of spatial point processes with a density
11.1.4 A useful coupling construction in the locally stable
and constant death rate case
11.1.5 Ergodic averages for spatial birth-death processes
11.2 Perfect simulation
11.2.1 General CFTP algorithms
11.2.2 Propp-Wilson’s CFTP algorithm
11.2.3 Propp-Wilson’s monotone CFTP algorithm
11.2.4 Perfect simulation of continuum Ising models
11.2.5 Read-once algorithm
11.2.6 Dominated CFTP
11.2.7 Clans of ancestors
11.2.8 Empirical findings
11.2.9 Other perfect simulation algorithms
App endices
A History, bibliography, and software

A.1 Brief history
A.2 Brief bibliography
A.3 Software
B Measure theoretical details

B.1 Preliminaries
B.2 Formal definition of point processes
B.3 Some useful conditions and results

C Moment measures and Palm distributions
C.1 Moment measures
C.1.1 Moment measures in a general setting
C.1.2 The second order reduced moment measure
C.2 Campbell measures and Palm distributions
C.2.1 Campbell measures and Palm distributions in a
general setting
C.2.2 Palm distributions in the stationary case
C.2.3 Interpretation of K and G as Palm expectations
D Perfect simulation of SNCPs
E Simulation of Gaussian fields
F Nearest-neighb our Markov p oint pro cesses

F.1 Definition and characterisation
F.2 Examples
F.3 Connected component Markov point processes
G Results for spatial birth-death pro cesses

G.1 Jump processes
G.2 Coupling constructions
G.3 Detailed balance
G.4 Ergodicity properties
References

Preface
Spatial point processes are used to model point patterns where the points
typically are positions or centres of objects in a two- or three-dimensional
region. The points may be decorated with marks (such as sizes or types
of the objects) whereby marked point processes are obtained. The areas
of applications are manifold and include astronomy, ecology, forestry,
geography, image analysis, and spatial epidemiology. For more than 30
years spatial point processes have been a major area of research in spa-
tial statistics, see e.g. Ripley (1977, 1981), Diggle (1983), Cressie (1993),
Stoyan & Stoyan (1994), Stoyan, Kendall & Mecke (1995), and van
Lieshout (2000). We expect that research in spatial point processes will
continue to be of importance as new technology makes huge amounts of
spatial point process data available and new applications emerge.
Some of the earliest applications of computational methods in statis-
tics are related to spatial point processes, cf. the historical account in
Section A.1. In the last decade computational methods, and particu-
larly Markov Chain Monte Carlo (MCMC) methods, have undergone
major developments. However, general textbooks on MCMC and sta-
tistical applications contain very little material associated with spatial
point processes. The recent survey papers by Geyer (1999) and Møller
& Waagepetersen (2003) and the book by van Lieshout (2000) contain
material on MCMC methods for spatial point processes, but a more com-
prehensive book which collects and unifies recent theoretical advances
and shows examples of applications in simulation-based inference for
spatial point process seems missing.
This book aims at filling this gap; it is concerned with simulation-
based inference for spatial point processes, with an emphasis on MCMC
methods. It should be accessible for a fairly general readership interested
in spatial models, including senior undergraduate students and Ph.D.
students in statistics, experienced statisticians, and applied probabilists.
Moreover, researchers working with complex stochastic systems in e.g.
environmental statistics, ecology, and materials science may benefit from
reading the parts of the book concerned with the statistical aspects.
A substantial part of the book consists of statistical methodology,
simulation studies, and examples of statistical analysis of the different
datasets introduced in Chapter 1. The book provides further a detailed

treatment of the mathematical theory for spatial point processes and
simulation-based methods. We have sought to make the book as self-
contained as possible, and give proofs of almost all mathematical results
in the book. Thereby the reader can be acquainted with the basic mathe-
matical techniques for point processes. The general mathematical theory
for point processes is heavily based on measure theory. In order to make
the book accessible to a wide audience, we mainly restrict attention to
point processes on Rd , whereby the mathematical treatment is simpli-
fied. However, most results generalise in an obvious way to more general
state spaces. Measure theoretical details are confined to Appendices B–
C. Readers with no or very limited knowledge on measure theory (or
who are less interested in the measure theoretical details) can read the
book without consulting these appendices, since only a very few proofs
(marked with †) rely on such knowledge.
The book is organised as follows. The first part (Chapters 1–4) deals
with background material for point processes, including Poisson point
processes and nonparametric methods based on summary statistics. The
second part (Chapters 5–6) concerns Cox and Markov point processes
which we consider the most useful model classes in data analysis. The
third part (Chapters 7–9) treats both some general background material
on MCMC methods, particularly Monte Carlo methods related to sta-
tistical inference as well as theoretical issues such as stability properties
of Markov chains, and simulation-based inference for Cox and Markov
point process models. The fourth part (Chapter 11 – Appendix E) deals
with some more specialised topics and technical issues, including recent
advances of perfect simulation for spatial point processes. Finally, Ap-
pendices A–G provide a brief history and bibliography, some comments
on software, and various technical topics.

Acknowledgments
We have been supported by the European Union’s TMR network “Sta-

tistical and Computational Methods for the Analysis of Spatial Data.
ERB-FMRX-CT96-0095,” by the Centre for Mathematical Physics and
Stochastics (MaPhySto), funded by a grant from the Danish National
Research Foundation, and by the Danish Natural Science Research Coun-
cil. We are grateful to Adrian J. Baddeley, Kasper K. Berthelsen, Anders
Brix, Peter J. Diggle, Marie-Colette van Lieshout, Shigeru Mase, Linda
Stougaard Nielsen, Antti Penttinen, Jakob Gulddahl Rasmussen, Diet-
rich Stoyan, and Robert L. Wolpert for their useful comments and as-
sistance with computer code and data. Our exposition of the subjects of
spatial point processes and simulation-based inference has further been
much inspired by Julian Besag, Charles J. Geyer, Eva Vedel Jensen, Wil-
frid S. Kendall, and Gareth Roberts. Our absorption in the work on the
book has demanded much patience and tolerance from our families to
whom we owe a special thank you. In particular, we dedicate this book
to Søren, Peter, and Katrine.
Jesper Møller
Rasmus Plenge Waagepetersen
Aalborg, Denmark.

CHAPTER 1
Examples of spatial point patterns
This chapter introduces some examples of spatial point patterns which

are used throughout this book for illustrative purposes. They have been
chosen both because of simplicity and because they are flexible enough to
show Markov chain Monte Carlo (MCMC) applications of many point
process models. Other examples of applications can be found in the
references given in this book, particularly many of those mentioned in
Section A.2.
Example 1.1 (Hickory trees) Figure 1.1 shows the positions of 85 hick-
ory trees in an 120 by 120m square subset of the Bormann research plot
in Duke Forest in Durham, North Carolina. This data set which served
as an example in Wolpert & Ickstadt (1998) is a subset of a much larger
data set including 7859 trees of 38 species, which was initially collected
in 1951–52 by Bormann and updated in 1974, 1982, 1989, and 1993 by N.
Christensen, R. Peet and members of the Duke University and University
of North Carolina botany departments in a continuing study of forest
maturation (Bormann 1953, Christensen 1977). The trees in Figure 1.1
seem to aggregate, possibly due to environmental conditions which we
later model by an unobserved intensity process (Chapters 5 and 10).
Example 1.2 (Amacrine cells) The points in Figure 1.2 show amacrine
cell centres observed in a rectangular area within the retinal ganglion
cell layer of a rabbit. There are 152 “on” cells and 142 “off” cells, corre-
sponding to cells which process “light-on” and “light-off” information.
The observation window is scaled to [0, 1.6065] × [0, 1]. Compared to the
point pattern in Figure 1.1, the two point patterns in Figure 1.2 seem
more regular, possibly due to a repulsive behaviour between the points.
Scientific hypotheses of interest are whether the two types of points can
be regarded as realisations of independent point processes, and whether
the division of cells into “on” respective “off” cells is completely random
given the positions of the cells. For details, see Wässle, Boycott & Illing
(1981) and Diggle (1986).
The amacrine cell data is an example of a multitype (or bivariate)
point pattern, and it is analysed in Diggle (1986) and van Lieshout &
Baddeley (1999) using nonparametric summary statistics (Chapter 4).

The univariate pattern of the “on” cell centres is analysed in several
papers, including Diggle & Gratton (1984) who use an ad hoc method
for estimation, Ogata & Tanemura (1984) and Särkkä (1993) who use
likelihood and pseudo likelihood methods (Chapter 9), and Heikkinen &
Penttinen (1999) and Berthelsen & Møller (2003) who use a nonpara-
metric Bayesian MCMC setting (Chapter 9).
Example 1.3 (Mucous membrane cells) The point pattern in Figure

1.3 shows cell centres in a cross section of the mucous membrane of
the stomach of a rat. The rectangular observation window is scaled to
[0, 1]× [0, 0.7]. The cells are divided into two classes: ECL cells and other
• • •
•
•• • ••• •
•• • •• • • •• • •
• • •• ••• •
•• •
• • • • •
• • • •
•
•• • • ••
• •
• •
•
•
• •
•
•
• •
•
•
• •
• •
• •
• •
• •
• • • ••
• • •
Figure 1.1 Positions of hickory trees.
Figure 1.2 Amacrine “on” (crosses) and “off ” (circles) cell centres.
©2004 by Chapman & Hall/CRC

cells for which, respectively, 86 and 807 cell centres are observed. An
additional feature compared with Example 1.2 is the increasing trend
in the intensity from top to bottom so that more points occur in the
bottom part of the window. A hypothesis of interest is whether the
spatially varying intensities of ECL cells and other cells are proportional.
A similar data set is examined in Nielsen (2000), using a particular
type of inhomogeneous Markov point process obtained by transforming
a homogeneous Markov point process (Chapters 6 and 9).
Figure 1.3 Mucous membrane ECL (crosses) and other (circles) cell centres.
Example 1.4 (Norwegian spruces) Figure 1.4 shows 134 discs, where
the centres are the positions of Norwegian spruces observed in a rect-
angular window of size 56 × 38m, and the radii are the stem diameters
multiplied by 5. As discussed in Penttinen, Stoyan & Henttonen (1992)
and Goulard, Särkkä & Grabarnik (1996) the “influence zone” of a tree
is about five times the stem diameter. The discs can be modelled by a
so-called marked point process (Section 2.2). The data has been analysed
in Fiksel (1984a), Penttinen et al. (1992), Stoyan et al. (1995), Goulard
et al. (1996), and Møller & Waagepetersen (2003) using different Markov
marked point process models (Chapters 6 and 9).
Example 1.5 (Weed) In this example the points are positions of 976
Trifolium spp. (clover) and 406 Veronica spp. (speedwell) weed plants
observed within 45 metal frames on a Danish barley field. The 45 frames
are of size 30×20cm and organised in 9 groups, each containing 5 frames,
where the vertical and horizontal distances between two neighbouring

. . . . . . . .
. .
. . . .
. . .
. . .
. . . . . .
. . . .
.
. .
. . .
. . . . . .
. . . .
. . .
. . . .
. .
. . . . . . . . . .
. . . . . . .
. . . . .
. . .
. . .
. . . . . . . . .
. .
. .
. . . . . .
. . . . . .
. . . . .
. .
. . . .
. . . .
. .
. .
. . . . . . .
Figure 1.4 Norwegian spruces. The radii of the discs equal 5 times the stem
diameters.
groups are 1 and 1.5m, respectively, see Figure 1.5. The size of the ex-
perimental area is 7.5 × 5m, where the longest side is parallel with the
ploughing direction. The weed plant positions are shown in Figure 1.6
where we have rotated the design 90◦ and omitted some space between
the frames. Note the trend in the point pattern of Trifolium spp. weed
plants: in general more plants occur in the upper frames, i.e. the frames
to the left in Figure 1.6.
5.0 m
7.5 m
Figure 1.5 Sampling design for weed data.

Figure 1.6 Positions of weed plants when the design is rotated 90◦ (some space
between the frames is omitted). Crosses correspond to Trifolium spp. and cir-
cles to Veronica spp.
This bivariate point pattern is a subset of a much larger dataset

where weed plants observed at eight different dates are considered (Fig-
ure 1.6 shows the observation at date 30 May 1996). The entire dataset
is analysed by different Cox process models (Chapters 5 and 10) in

Brix & Møller (2001) and Brix & Chadoeuf (2000). See also Møller &
Waagepetersen (2003).

CHAPTER 2
Introduction to point processes
This chapter gives an informal introduction to point process and to

marked point processes with points in Rd (the d-dimensional Euclid-
ian space). It also states the definitions and notation used throughout
the book. The rigorous mathematical framework for point processes on
a general state space is given in Appendix B, which can be skipped by
readers who are unfamiliar with or less interested in measure theoretical
details. Section A.1 describes the history of spatial point processes, and
Section A.2 contains several references for further reading.
2.1 Point pro cesses on Rd
A spatial point process X is a random countable subset of a space S.

Throughout this book, unless otherwise stated, we will always assume
that S ⊆ Rd . Often S will be a d-dimensional box or the whole of Rd ,
but it could also be e.g. the (d − 1)-dimensional unit sphere. In practice
we observe only the points contained in a bounded observation window
W ⊆ S. In Examples 1.2–1.4, d = 2 and W is either a rectangle or a
union of rectangles.
We restrict attention to point processes X whose realisations are lo-
cally finite subsets of S. Formally, for any subset x ⊆ S, let n(x) denote
the cardinality of x, setting n(x) = ∞ if x is not finite. Then x is said
to be locally finite, if n(xB ) < ∞ whenever B ⊆ S is bounded, where
xB = x ∩ B
is the restriction of a point configuration x to B (similarly XB is the

restriction of X to B). Thus X takes values in the space defined by
Nlf = {x ⊆ S : n(xB ) < ∞ for all bounded B ⊆ S}.
Elements of Nlf are called locally finite point configurations, and they
will be denoted by x, y, . . ., while ξ, η, . . . will denote points in S. The
empty point configuration is denoted ∅. We shall abuse notation and
write x ∪ ξ for x ∪ {ξ}, x \ η for x \ {η}, etc., when x ∈ Nlf and ξ, η ∈ S.

2.2 Marked p oint pro cesses and multivariate p oint pro cesses
Let Y be a point process on T ⊆ Rd (the reason for shift in notation
will become apparent in Section 2.3). Given some space M , if a random
“mark” mξ ∈ M is attached to each point ξ ∈ Y , then
X = {(ξ, mξ ) : ξ ∈ Y }
is called a marked point process with points in T and mark space M . In
this book we mainly consider the cases where M is either a finite set or
a subset of Rp , p ≥ 1. For more general cases of marked point processes,
see Stoyan & Stoyan (1994), Schlather (2001), and the references therein.
For the disc process in Example 1.4, M = (0, ∞), where (ξ, mξ ) is
identified with the disc with centre ξ and radius mξ . Similarly, we may
obtain marked point processes with other kinds of geometric objects (line
segments, ellipses, etc.) which can be identified with points in Rp . Such
processes are called germ-grain models where ξ (the germ) specifies the
location of the object mξ (the grain).
Another simple example is a multitype point process, where M =
{1, . . . , k} and the marks specify k different types of points (e.g. differ-
ent types of cells or weed plants as in Examples 1.2, 1.3 , and 1.5 where
k = 2). This is equivalent to a k-dimensional multivariate point process,
that is a tuple (X1 , . . . , Xk ) of point processes X1 , . . . , Xk corresponding
to the k different types of points.
2.3 Unified framework for p oint pro cesses and marked p oint
pro cesses
In this book we distinguish between point processes on S ⊆ Rd and
marked point processes with points in T ⊆ Rd . However, both types
of processes as well as point processes on non-Euclidian spaces can be
handled in a unified framework where S is a general metric space, see
Appendix B (for a marked point process as in Section 2.2, S = T × M ).
We denote the metric d(·, ·), and for ξ ∈ S and r ≥ 0, we let
b(ξ, r) = {η ∈ S : d(ξ, η) ≤ r}
denote the closed ball in S with centre ξ and radius r. Examples of
metrics are given in Appendix B. When S ⊆ Rd , we let always, unless
otherwise stated, d(ξ, η) = ξ − η be the usual Euclidian distance.
For the mathematical treatment of point processes we equip S and Nlf
with appropriate σ-algebras denoted B and Nlf , respectively (for details,
see Section B.2). Readers unfamiliar with measure theory may just think
of B and Nlf as very large families of subsets of S and Nlf . We shall often
just write B ⊆ S and F ⊆ Nlf when we mean B ∈ B and F ∈ Nlf .
Appendix B contains some useful characterisation and uniqueness re-

sults for spatial point processes in a general setting. An informal account
is given below when S ⊆ Rd . We also briefly explain how a technique
called the standard proof becomes useful.
2.3.1 Characterisation using void events

Let
B0 = {B ∈ B : B is bounded}.
For a point process X on S consider the count function
N (B) = n(XB )
which is the random number of points falling in B ⊆ S. Sets of the form
FB = {x ∈ Nlf : n(xB ) = 0} with B ∈ B0 are called void events. Note
that X ∈ FB if and only if N (B) = 0. Under mild conditions (which
will be satisfied throughout the book, cf. Theorem B.1 in Section B.3)
the distribution of X (i.e. P (X ∈ F ), F ∈ Nlf ) is determined by its void
probabilities defined by
v(B) = P (N (B) = 0), B ∈ B0 .
Similarly, the distribution of a multivariate point process (X1 , . . . , Xk )
with X1 , . . . , Xk defined on T ⊆ Rd and corresponding count functions
N1 , . . . , Nk , is determined by the void probabilities
v(B1 , . . . , Bk ) = P (N1 (B1 ) = 0, . . . , Nk (Bk ) = 0)
for bounded B1 , . . . , Bk ⊆ T .
Consider

ν(E) = E 1[(ξ, X) ∈ E], E ⊆ S × Nlf ,
ξ∈X
where 1[·] denotes the indicator function. Then ν is a measure, mean-

∞
ing that ν(·) ≥ 0, ν(∅) = 0, and ν(∪∞ E
i=1 i ) = i=1 ν(Ei ) for disjoint
E1 , E2 , . . . ⊆ S × Nlf . In fact ν is uniquely determined by its values on
sets of the form E = B × F where B ∈ B0 and F is a void event. Thus
ν is an example of a measure defined in terms of a certain expectation,
where the measure is determined by its values on a small class of sets.
We shall often consider such measures and characterisations, which are
treated in more detail in Appendix B.
2.3.2 Characterisation using the generating functional

The generating functional for a point process X on S is defined by

GX (u) = E u(ξ) (2.1)
ξ∈X

for functions u : S → [0, 1] with {ξ ∈ S : u(ξ) < 1} bounded. For B ∈ B0
and u(ξ) = t1[ξ∈B] with 0 ≤ t ≤ 1, GX (u) = EtN (B) is the generating
function for N (B). Hence the distribution of X is uniquely determined
by GX .
The generating functional can be a very useful tool, see e.g. Daley &
Vere-Jones (1988), but we shall only use it a few times in this book. By
putting additional assumptions on X, GX (u) can be defined for more
general functions u, see Daley & Vere-Jones (1988) and Proposition 3.3
in Section 3.1.2.
2.3.3 The standard proof
Suppose we are given two expressions in terms of integrals or expecta-

tions of an arbitrary nonnegative function h defined on some space H and
that we want to establish the equality of these
expressions. For
instance,
we shall later consider the expressions E ξ∈X h(ξ) and h(ξ)ρ(ξ)dξ
where H = S and ρ : S → [0, ∞) is a certain function called the inten-
sity function. Assume that the two expressions become measures when-
ever h is an indicator function; e.g. both ν1 (A) = E ξ∈X 1[ξ ∈ A] and

ν2 (A) = A ρ(ξ)dξ are measures for A ⊆ S. Then it suffices to verify the
equality for every indicator function (for more details see the remark to
Theorem 16.11 in Billingsley 1995). We shall refer to this technique as
the standard proof.
2.4 Space-time processes
A spatial point pattern observed within some bounded window W ⊆

Rd is usually a realisation of a continuous-time process observed at a
fixed time. Sometimes point patterns are observed at discrete times for
a continuous-time process. One example is the weed data (Example 1.5)
where the full space-time dataset is analysed in Brix & Møller (2001).
Other examples include the space-time modelling of localised cases of
gastro-intestinal infections (Brix & Diggle 2001), forest trees (Rathbun
& Cressie 1994b), geological data (Fiksel 1984b), sand dunes (Møller &
Sørensen 1994), and earthquake occurrences (Ogata 1998, Zhung, Ogata
& Vere-Jones 2002). See also Schoenberg, Brillinger & Guttorp (2002)
and the references therein.
For many applications only one realisation of a spatial point process is
available and information concerning the underlying space-time process
is missing. This is also the case for the statistical analyses considered
in this book, and the theory we present will accordingly be restricted
to this situation. Statistical inference for space-time processes is often

simpler than for spatial point processes, see Daley & Vere-Jones (1988,
2003) and Section 9.2.5.
Space-time processes and Monte Carlo methods provide indispensable
tools for simulation-based inference of spatial point process models. For
example, even for the Poisson process, which is a simple and fundamental
point process model studied in detail in Chapter 3, most statistics have
intractable or unknown distributions. The Poisson point process (and
some related point process models) can easily be simulated, but most
point process models are so complicated that direct simulation is infea-
sible. Instead simulations can be obtained from discrete or continuous
time Markov chains with a specified stationary distribution. Chapters 7–
11 treat these points in detail.

CHAPTER 3
Poisson point processes
Poisson point processes play a fundamental role. They serve as a tractable

model class for “no interaction” or “complete spatial randomness” in
spatial point patterns. They also serve as reference processes when sum-
mary statistics are studied (Chapter 4) and when more advanced point
process models are constructed (Chapters 5–6).
In this chapter we study some of the most important properties of
Poisson processes. Section 3.1 contains basic definitions and results,
particularly concerning existence and construction of Poisson processes.
Section 3.2 contains some further important results, including the use-
ful Slivnyak-Mecke theorem and results for superpositioning, thinning,
simulation, and densities for Poisson processes. Section 3.3 deals with
marked Poisson processes.
Further material on Poisson processes can be found in Daley & Vere-
Jones (1988), Kingman (1993), and Stoyan et al. (1995).
3.1.1 Definitions
We start by considering Poisson point processes defined on a space S ⊆

Rd and specified by a so-called
intensity function ρ : S → [0, ∞) which
is locally integrable, i.e. B ρ(ξ)dξ < ∞ for all bounded B ⊆ S. This is
by far the most important case for applications.
Remark 3.1 In the definition below of a Poisson process we use only

the intensity measure µ given by

µ(B) = ρ(ξ)dξ, B ⊆ S. (3.1)
B
This measure is locally finite, i.e. µ(B) < ∞ for bounded B ⊆ S, and
diffuse, i.e. µ({ξ}) = 0 for ξ ∈ S. For readers familiar with measure
theory it should be obvious how many of the definitions and results in
this chapter extend to a general metric space S as in Appendix B and a
locally finite diffuse measure µ on S.

Before defining a Poisson process with intensity function ρ, we define
a related process:
Definition 3.1 Let f be a density function on a set B ⊆ S, and let

n ∈ N (where N = {1, 2, 3, . . .}). A point process X consisting of n i.i.d.
points with density f is called a binomial point process of n points in
B with density f . We write X ∼ binomial(B, n, f ) (where ∼ means
“distributed as”).

Note that B in Definition 3.1 has volume |B| > 0, since B f (ξ)dξ = 1.
In the simplest case, |B| < ∞ and each of the n i.i.d. points follows
Uniform(B), the uniform distribution on B, i.e. f (ξ) = 1/|B| is the
uniform density on B.
Definition 3.2 A point process X on S is a Poisson point process with

intensity function ρ if the following properties are satisfied (where µ is
given by (3.1)):
(i) for any B ⊆ S with µ(B) < ∞, N (B) ∼ po(µ(B)), the Poisson
distribution with mean µ(B) (if µ(B) = 0 then N (B) = 0);
(ii) for any n ∈ N and B ⊆ S with 0 < µ(B) < ∞, conditional on
N (B) = n, XB ∼ binomial(B, n, f ) with f (ξ) = ρ(ξ)/µ(B).
We then write X ∼ Poisson(S, ρ).
For any bounded B ⊆ S, µ determines the expected number of points

in B,
EN (B) = µ(B).
Heuristically, ρ(ξ)dξ is the probability for the occurrence of a point in
an infinitesimally small ball with centre ξ and volume dξ.
Definition 3.3 If ρ is constant, the process Poisson(S, ρ) is called a

homogeneous Poisson process on S with rate or intensity ρ; else it is said
to be an inhomogeneous Poisson process on S. Moreover, Poisson(S, 1)
is called the standard Poisson point process or unit rate Poisson process
on S.
Examples of simulated homogeneous and inhomogeneous Poisson point

processes are shown in Figure 3.1. For a homogeneous Poisson process
on Rd , ρ−1/d is a scale parameter, since X ∼ Poisson(Rd , 1) implies that
{ρ−1/d ξ : ξ ∈ X} ∼ Poisson(Rd , ρ).
A homogeneous Poisson point process is both stationary and isotropic
in the following sense.
Definition 3.4 A point process X on Rd is stationary if its distribution

• • •• •
• ••
•• • • • •
• • • • • •
• • • • ••
• •
• • • • •
• • •
• • •• • •
• • • •••
•• •• •• • •••
•• •
•
• • • • • • •
• • • • •
• •• • • •
• • •• • • •
• • • •
• •
• • •
• • • •
• • • • •
• • •
•• •• • •
•• • • • • •
• • • • • • •
•• •• • • • •
• •
• •
• •
• • • •
• •
• • • •• •
• •• • • • • • •• • • • •• • •• • •
• •
•
•
• • • • •• • • •
• • •• • •• • • • •• • ••
• • • • • • • • •• •
•• • • • •• • • • • •• •
• • • • •• •• ••• • ••• • • • • •••• ••
• • •
• • •• • •• • •• • • • • • • • • • • •
•• • • • • • •• ••
••
• •• •• • •• • • •• • •
• • • • • ••• •• • • • • •• ••• •• • • •• • • • • • • • •• •• • • •• •
Figure 3.1 Simulation of homogeneous (left) and inhomogeneous (right) Pois-

son processes on S = [0, 1] × [0, 0.7]. In both cases the expected num-
ber of points is 150. For the inhomogeneous Poisson process, ρ(ξ1 , ξ2 ) ∝
exp(−10.6ξ2 ), (ξ1 , ξ2 ) ∈ S.
is invariant under translations, that is, the distribution of X +s = {ξ+s :

ξ ∈ X} is the same as that of X for any s ∈ Rd . It is isotropic if its
distribution is invariant under rotations about the origin in Rd , i.e. the
distribution of OX = {Oξ : ξ ∈ X} is the same as that of X for any
rotation O around the origin.
3.1.2 Existence and independent scattering property

We shall often use the following expansion for the Poisson process:
Proposition 3.1
(i) X ∼ Poisson(S, ρ) if and only if for all B ⊆ S
with µ(B) = S ρ(ξ)dξ < ∞ and all F ⊆ Nlf ,
P (XB ∈ F )
∞
n
exp(−µ(B))
= ··· 1[{x1 , . . . , xn } ∈ F ] ρ(xi )dx1 · · · dxn
n=0
n! B B i=1
(3.2)
where the integral for n = 0 is read as 1[∅ ∈ F ].
(ii) If X ∼ Poisson(S, ρ), then for functions h : Nlf → [0, ∞) and B ⊆ S
with µ(B) < ∞,

E h(XB )
∞ n
exp(−µ(B))
= ··· h({x1 , . . . , xn }) ρ(xi )dx1 · · · dxn .
n=0
n! B B i=1
(3.3)
Proof. (i) follows immediately from (i) and (ii) in Definition 3.2. (i) and
the standard proof imply (ii).

If S is bounded (or just S ρ(ξ)dξ < ∞), the existence of Poisson(S, ρ)
follows easily from Proposition 3.1 (with B = S): let n(X) ∼ po(µ(S))
and X|n(X) = n ∼ binomial (S, n, ρ/µ(S)) for n ∈ N. For an unbounded
S, existence of Poisson(S, ρ) is verified in the following theorem.
Theorem 3.1 X ∼ Poisson(S, ρ) exists and is uniquely determined by

its void probabilities
v(B) = exp(−µ(B)), bounded B ⊆ S. (3.4)
Proof. Let ξ ∈ S be an arbitrary point and set Bi = {η ∈ S : i − 1 ≤

η−ξ < i} for i ∈ N. Clearly, S is a disjoint union of the bounded Bi . Let
X = ∪∞ 1 Xi where Xi ∼ Poisson(Bi , ρi ), i = 1, 2, . . . , are independent,
and where ρi is the restriction of ρ to Bi . Then for bounded B ⊆ S,
∞
∞

P (X ∩ B = ∅) = P (Xi ∩ B = ∅) = exp(−µ(B ∩ Bi ))
i=1 i=1
∞

= exp − µ(B ∩ Bi ) = exp(−µ(B)).
i=1
This is the void probability for a Poisson process with intensity measure
µ, so the existence and uniqueness follow from Section 2.3 (or Theo-
rem B.1 in Section B.3).
Remark 3.2 In fact, “bounded” in (3.4) can be replaced by “compact”

(Matheron 1975) or by “finite union of rectangles” (Rényi’s theorem, see
e.g. Kingman 1993). Theorem 3.1 extends immediately to independent
Poisson processes: if X1 and X2 are independent Poisson processes on
S with intensity measures µ1 and µ2 , the distribution of (X1 , X2 ) is
uniquely determined by the void probabilities
P (X1 ∩ A = ∅, X2 ∩ B = ∅) = exp(−µ1 (A) − µ2 (B)) (3.5)
for bounded A, B ⊆ S.
By Theorem 3.1, Definition 3.2 states more than needed for charac-
terising a Poisson process. However, as seen in the proof of Theorem 3.1
and many places in the sequel, (i) and (ii) in Definition 3.2 provide an
easy construction of a Poisson process. Sometimes in the literature, (ii)
is replaced by the condition that N (B1 ), . . . , N (Bn ) are independent for
disjoint sets B1 , . . . , Bn ⊆ S and n ≥ 2. This property is called indepen-
dent scattering. It can be extended as follows.
Proposition 3.2 If X is a Poisson process on S, then XB1 , XB2 , . . . are

independent for disjoint sets B1 , B2 , . . . ⊆ S.

Proof. We need only to verify that XB1 , . . . , XBn are independent for
disjoint sets B1 , . . . , Bn ⊆ S and n ≥ 2. This follows straightforwardly
from Proposition 3.1 when n = 2 and B1 , B2 ⊆ S are bounded and dis-
joint (with B = B1 ∪B2 ), and hence by induction also when B1 , . . . , Bn ⊆
S are bounded and disjoint. Hence XB1 , XB2 , . . . are independent for
bounded disjoint Bi ⊆ S, i = 1, 2, . . .. Since subsets of S are countable
unions of bounded subsets of S, the assertion is now seen to hold for any
disjoint B1 , . . . , Bn ⊆ S and n ∈ N.
Proposition 3.2 explains the terminology of no interaction and com-

plete spatial randomness in the Poisson process.
We conclude by deriving the generating functional for a Poisson pro-
cess, noticing that in this case the definition (2.1) of GX (u) extends to
any function u : S → [0, 1], setting exp(−∞) = 0.
Proposition 3.3 If X ∼ Poisson(S, ρ) then

GX (u) = exp − (1 − u(ξ))ρ(ξ)dξ (3.6)
S
for functions u : S → [0, 1].

∞
Proof. We verify this only in the special case when u(ξ) = 1 − i=1 (1 −
ai )1[ξ ∈ Ci ], where 0 ≤ ai ≤ 1 and the Ci ⊆ S are bounded and pair-
wise disjoint. (The general case follows then by standard arguments and
the monotone convergence theorem from measure theory.) By Proposi-
tion 3.2, N (Ci ) ∼ po(µ(Ci )), i = 1, 2, . . ., are independent. Hence,
∞
∞
∞

N (Ci ) N (Ci )
GX (u) = E ai = Eai = exp(−(1 − ai )µ(Ci ))
i=1 i=1 i=1
∞

= exp − (1 − ai )µ(Ci ) = exp − (1 − u(ξ))ρ(ξ)dξ .
i=1
3.1.3 Constructions of stationary Poisson processes

A construction of a Poisson process with a given intensity function is
given in the proof of Theorem 3.1. In the stationary case Quine & Watson
(1984) give an easier construction using a shift to polar coordinates. This
construction is given in the following proposition, letting
ωd = π d/2 /Γ(1 + d/2), σd = 2π d/2 /Γ(d/2),
denote the volume and surface area of the d-dimensional unit ball.

Proposition 3.4 Let S1 , U1 , S2 , U2 , . . . be mutually independent where
each Ui is uniformly distributed on {u ∈ Rd : u = 1} (the (d −
1)-dimensional unit sphere), and each Si ∼ Exp(ρωd ) is exponentially
distributed with mean 1/(ρωd) for a ρ > 0. Let R0 = 0 and Rid =
d
Ri−1 + Si , i = 1, 2, . . .. Then X = {R1 U1 , R2 U2 , . . .} ∼ Poisson(Rd , ρ).
Proof. By Theorem 3.1 we have to verify that v(B) = exp(−ρ|B|) for

any bounded B ⊂ Rd . This follows if for any r > 0, (i) N (b(0, r)) ∼
po(ρωd rd ) and (ii) conditional on N (b(0, r)) = n, we have that ξ1 , . . . , ξn
are distributed as n uniform points on b(0, r) ordered according to Eu-
clidian distance. Because if B ⊆ b(0, r) then by (i) and (ii),
P (N (B) = 0)
∞

= P (N (B) = 0|N (b(0, r) = n)P (N (b(0, r) = n)
n=0
∞
= (|b(0, r) \ B|/|b(0, r)|)n exp(−ρ|b(0, r)|)(ρ|b(0, r)|)n /n!
n=0
= exp(ρ|b(0, r) \ B| − ρ|b(0, r)|) = exp(−ρ|B|).
We verify now (i) and (ii).† For a fixed n ∈ N, the joint density of
(S1 , U1 , . . . , Sn , Un ) is
g1 (s1 , u1 , . . . , sn , un ) = (ρωd )n exp(−ρωd rnd )σd−n

n
where rn = ( j=1 sj )1/d and the density is with respect to the n-fold
product measure of Lebesgue measure on (0, ∞) times surface measure
on the d-dimensional unit ball. The conditional probability of N (b(0, r)) =
n given (S1 , U1 , . . . , Sn , Un ) = (s1 , u1 , . . . , sn , un ) is
1[rn ≤ r] exp(−ρωd (rd − rnd )),
since N (b(0, r)) = n is equivalent to Rn ≤ r and Sn+1 > rd − Rnd . The

joint density of (S1 , U1 , . . . , Sn , Un , 1[N (b(0, r)) = n]) therefore satisfies
g2 (s1 , u1 , . . . , sn , un , 1) = 1[rn ≤ r](ρωd )n σd−n exp(−ρωd rd ).
Integrating g2 (s1 , u1 , . . . , sn , un , 1) over (s1 , u1 , . . . , sn , un ) we obtain
P (N (b(0, r)) = n) = exp(ρωd rd )(ρωd rd )n /n!,
and so (i) is verified. Thus the conditional density of (R1 , U1 , . . . , Rn , Un )
† Readers who are unfamiliar with measure theory may prefer to skip this part of
the proof.

given N (b(0, r)) = n is
g3 (r1 , u1 , . . . , rn , un ) =

n
n!σd−n r−nd 1[0 ≤ r1 ≤ r2 ≤ · · · ≤ rn ≤ r] drid−1
i=1
d d
where σd r /d = ωd r is the volume of b(0, r). Transforming from polar
coordinates (Ri , Ui ) to Xi = Ri Ui (see e.g. Proposition 2.8 in Jensen
1998) we further obtain that the conditional density of (ξ1 , . . . , ξn ) given
N (b(0, r)) = n is
g4 (ξ1 , . . . , ξn ) = n!(ωd rd )−n 1[ξ1 ≤ ξ2 ≤ · · · ≤ ξn ≤ r].
Thereby (ii) is verified.
When d = 1 it seems more natural to apply the following proposition

where points on the real line are ordered as usual.
Proposition 3.5 Let . . . , T−2 , T−1 , T1 , T2 , . . . be mutually independent

and Exp(ρ)-distributed for a ρ > 0. Let ξ0 = 0, and for i = 1, 2, . . ., let
ξi = ξi−1 +Ti and ξ−i = ξ−i+1 −T−i . Then X = {. . . , ξ−2 , ξ−1 , ξ1 , ξ2 , . . .}
∼ Poisson(R, ρ).
Proof. This is verified along similar lines as in the proof of Proposi-

tion 3.4.
Remark 3.3 The point ξ0 = 0 serves as a starting point when gen-

erating ξ1 , ξ2 , . . . and ξ−1 , ξ−2 , . . .. By stationarity, an arbitrary point
ξ0 ∈ Rd can be chosen. Note that the exponential distribution has no
memory in the sense that for a > 0 and t > 0,
T ∼ Exp(a) ⇒ T − t|T > t ∼ Exp(a).
So the construction of Poisson(I, ρ) on an interval I = (a, b), where
either a or b or both are finite, is similar to that of Proposition 3.5,
taking either ξ0 = a or ξ0 = b.
The interval [ξ−1 , ξ1 ] which contains ξ0 is longer than a typical in-
terval [ξi−1 , ξi ]. Specifically, the length of a typical interval is Exp(ρ)-
distributed, while ξ1 − ξ−1 is distributed as the sum of two indepen-
dent Exp(ρ)-distributed random variables. This seemingly “paradox” is
due to sampling bias: Consider a stationary point process with points
. . . < η−1 < η0 < η1 < . . . on the real line and which is independent
of a random real variable Z, and suppose that we sample the interval
[ηk−1 , ηk ] which contains Z. Let L0 denote the length of this “point
sampled” interval and let L be the length of a typical interval (i.e. the
distribution of L is given by the so-called Palm distribution, see e.g.

Stoyan et al. 1995 and Appendix C). Since larger intervals are more
likely to get sampled, L0 will on the average be greater than L. In fact,
it is not hard to verify that for functions k : (0, ∞) → (0, ∞),
Ek(L)/EL = E(k(L0 )/L0 ),
so
EL0 = E(L2 )/EL ≥ (EL)2 /EL = EL.
Even a stronger result holds, namely P (L > l) ≤ P (L0 > l) for all t ≥ 0
(Mecke 1999).
Remark 3.4 Proposition 3.5 extends to the inhomogeneous case by

transforming a standard Poisson process. For simplicity suppose that
ρ(ξ) > 0 for all ξ ∈ R and that
∞
H(t) = ρ(s)ds
t
is finite for all t ∈ R. Then H(t) is strictly decreasing. Let X = {H −1 (η) :

η ∈ Y } where Y ∼ Poisson(R, 1). Then X ∼ Poisson(R, ρ), since

P (X∩A = ∅) = P (Y ∩H(A) = ∅) = exp(−|H(A)|) = exp − ρ(ξ)dξ
A
for bounded A ⊂ R. This construction is easily modified to the case

where ρ(ξ) ≥ 0.
For a fixed ξ0 ∈ Rd , enumerate the points in Y as · · · > η2 > η1 >
η−1 > η−2 > · · · so that η1 ≥ η0 > η−1 where η0 = H(ξ0 ), and let ξi =
H −1 (η−i ). Then the differences Ti = ξi −ξi−1 and T−i = ξ−i+1 −ξ−i , i =
1, 2, . . ., are mutually independent but not exponentially distributed (un-
less of course ρ(·) is constant). The sequence . . . , ξ−2 , ξ−1 , ξ0 , ξ1 , ξ2 , . . .
is a Markov chain which can be viewed as the arrival times of a birth
process with birth rate ρ.
3.2 Further results

Poisson processes are very tractable for mathematical analysis. This sec-
tion collects several useful results.
3.2.1 Slivnyak-Mecke’s theorem

The simplest way of characterising a Poisson point process is by its void
probabilities (3.4). A less well known but very useful characterisation
of a Poisson process is provided by the following theorem called the
Slivnyak-Mecke theorem (Mecke 1967).

Theorem 3.2 If X ∼ Poisson(S, ρ), then for functions h : S × Nlf →
[0, ∞),

E h(ξ, X \ ξ) = Eh(ξ, X)ρ(ξ)dξ (3.7)
ξ∈X S
(where the left hand side is finite if and only if the right hand side is
finite).
Proof. Consider first the case where

h(ξ, X \ ξ)
ξ∈X

only depends on X through XB for some B ⊆ S with B ρ(ξ)dξ < ∞.
Applying Proposition 3.1 we obtain

E h(ξ, X \ ξ) =
ξ∈X
∞ n n
exp(−µ(B))
··· h(xi , {x1 , . . . , xn } \ xi ) ρ(xi )dx1 · · · dxn
n=1
n! B B i=1 i=1
∞ n
exp(−µ(B))
= · · · h(xn , {x1 , . . . , xn−1 }) ρ(xi )dx1 · · · dxn
n=1
(n − 1)! B B i=1

= Eh(ξ, X)ρ(ξ)dξ.
S
Consider next the general case.† If we can prove (3.7) when h(ξ, x) =
1[(ξ, x) ∈ F ], F ⊆ S × Nlf , is an indicator function, then by the standard
proof, (3.7) is satisfied for any nonnegative function h. The left and
right sides in (3.7) both become measures on S × Nlf when we consider
indicator functions h(ξ, x) = 1[(ξ, x) ∈ F ], F ⊆ S × Nlf , and they are
both equal to EN (B) = B ρ(ξ)dξ if h(ξ, x) = 1[(ξ, x) ∈ B × Nlf ] for
bounded B ⊆ S. By Lemma B.4 in Section B.3, we thus just need to
verify (3.7) when h is of the form
h(ξ, x) = 1[ξ ∈ C, n(xA ) = 0], bounded A, C ⊆ S. (3.8)
But this follows from the first part of the proof, since B = A ∪ C is
bounded and ρ is locally integrable.
By induction we easily obtain the following extended Slivnyak-Mecke

theorem.
† Readers who are unfamiliar with measure theory may prefer to skip this part of
the proof.

Theorem 3.3 If X ∼ Poisson(S, ρ), then for any n ∈ N and any func-
tion h : S n × Nlf → [0, ∞),
=

E h(ξ1 , . . . , ξn , X \ {ξ1 , . . . , ξn })
ξ1 ,...,ξn ∈X

n
= ··· Eh(ξ1 , . . . , ξn , X) ρ(ξi )dξ1 · · · dξn (3.9)
S S i=1
where the = over the summation sign means that the n points ξ1 , . . . , ξn
are pairwise distinct.
Proof. For n = 1, (3.7) and (3.9) obviously agree. For n ≥ 2, consider
=

h̃(ξ, x) = h(ξ, ξ2 , . . . , ξn , x \ {ξ2 , . . . , ξn }).
ξ2 ,...,ξn ∈x
By (3.7) and the induction hypothesis, the left hand side in (3.9) is equal
to

E h̃(ξ, X \ ξ) = Eh̃(ξ, X)ρ(ξ)dξ
ξ∈X S

= ··· Eh(ξ, ξ2 , . . . , ξn , X)ρ(ξ2 ) · · · ρ(ξn )dξ2 · · · dξn ρ(ξ)dξ
S S S
which agrees with the right side in (3.9).
3.2.2 Superpositioning and thinning

We now define two basic operations for point processes.
Definition 3.5 A disjoint union ∪∞

i=1 Xi of point processes X1 , X2 , . . .,
is called a superposition.
Definition 3.6 Let p : S → [0, 1] be a function and X a point process

on S. The point process Xthin ⊆ X obtained by including ξ ∈ X in Xthin
with probability p(ξ), where points are included/excluded independently
of each other, is said to be an independent thinning of X with retention
probabilities p(ξ), ξ ∈ S. Formally, we can set
Xthin = {ξ ∈ X : R(ξ) ≤ p(ξ)}
where R(ξ) ∼ Uniform[0, 1], ξ ∈ S, are mutually independent and inde-
pendent of X.
As shown in the following two propositions, the class of Poisson pro-
cesses is closed under both superpositioning and independent thinning.

Proposition 3.6 If X i ∼ Poisson(S, ρi ), i = 1, 2, . . . , are mutually
independent and ρ = ρi is locally integrable, then with probability
one, X = ∪∞
i=1 Xi is a disjoint union, and X ∼ Poisson(S, ρ).
Proof. The first part can easily be verified using Proposition 3.1 and
considering two independent Poisson processes within a bounded ball
B which expands to S; it is also verified in the Disjointness Lemma in
Kingman (1993). The second part follows from Theorem 3.1: for bounded
B ⊆ S,
∞
∞

P (XB = ∅) = P (Xi ∩ B = ∅) = exp(−µi (B)) = exp(−µ(B))
i=1 i=1

where µi (B) = B ρi (ξ)dξ.
Proposition 3.7 Suppose that X ∼ Poisson(S, ρ) is subject to inde-

pendent thinning with retention probabilities p(ξ), ξ ∈ S, and let
ρthin (ξ) = p(ξ)ρ(ξ), ξ ∈ S.
Then Xthin and X \ Xthin are independent Poisson processes with inten-
sity functions ρthin and ρ − ρthin , respectively.

Proof. Let µthin be given by µthin (B) = ρthin (ξ)dξ. By (3.5), we only
need to verify that
P (Xthin ∩A = ∅, (X \Xthin )∩B = ∅) = exp(−µthin (A)−µ(B)+µthin (B))
(3.10)
for bounded A, B ⊆ S. By Proposition 3.1 and Definition 3.6, for any
bounded C ⊆ S,
∞ n
1 −µ(C)
P (Xthin ∩ C = ∅) = e (1 − p(ξ))ρ(ξ)dξ
n=0
n! C
= exp(−µthin (C)).
By symmetry,
P ((X \ Xthin ) ∩ C = ∅) = exp(−(µ − µthin )(C)).
The result now follows by Proposition 3.2: for bounded A, B ⊆ S,
P (Xthin ∩ A = ∅, (X \ Xthin ) ∩ B = ∅)
=P (X ∩ A ∩ B = ∅)P (Xthin ∩ (A \ B) = ∅)P ((X \ Xthin ) ∩ (B \ A) = ∅)
= exp(−µ(A ∩ B) − µthin (A \ B) − (µ − µthin )(B \ A))
whereby (3.10) is obtained.
The following corollary shows that inhomogeneous Poisson processes
often can be derived by thinning a homogeneous Poisson process.

Corollary 3.1 Suppose that X ∼ Poisson(Rd , ρ) where the intensity
function ρ is bounded by a finite constant c. Then X is distributed as
an independent thinning of Poisson(Rd , c) with retention probabilities
p(ξ) = ρ(ξ)/c.
Proof. Follows immediately from Proposition 3.7.
3.2.3 Simulation of Poisson processes

Simulation of a Poisson point process X ∼ Poisson(Rd , ρ) within a
bounded set B ⊂ Rd is usually easy.
Consider first the homogeneous case where ρ(ξ) = ρ0 > 0 is constant
for all ξ ∈ B. For d = 1, if B is an interval, we may simulate the
restriction to B of the Poisson process considered in Proposition 3.5.
For d ≥ 2, we consider three cases:
1. If B = b(0, r) is a ball, simulation of XB is straightforward by Propo-
sition 3.4: simulate S1 , . . . , Sm and U1 , . . . , Um−1 , where m is given
by that Rm−1 ≤ r < Rm , and return XB = {R1 U1 , . . . , Rm−1 Um−1 }.
This radial simulation procedure is due to Quine & Watson (1984).
2. If B = [0, a1 ]×· · ·×[0, ad ] is a box, then by (i)–(ii) in Definition 3.2, we
may first generate the number of points N (B) ∼ po(ρ0 a1 · · · ad ) (see
e.g. Ripley, 1987), and second the locations of the N (B) independent
and uniformly distributed points in B. An alternative is to apply
Proposition 3.5, where we exploit that by (i)–(ii) in Definition 3.2,
• the point process consisting of the first coordinates ξ (1) of the
points ξ = (ξ (1) , ξ (2) , . . . , ξ (d) ) ∈ XB is Poisson([0, a1 ], ρ0 a2 · · · ad ),
• the remaining components (ξ (2) , . . . , ξ (d) ) of such points are inde-
pendent and uniformly distributed on [0, a2 ] × · · · × [0, ad ].
3. If B is bounded but neither a ball nor a box, we simply simulate X
on a ball or box B0 containing B, and disregard the points falling in
B0 \ B.
Suppose next that XB is inhomogeneous with ρ(ξ), ξ ∈ B, bounded by
a constant ρ0 > 0. Then we may first generate a homogeneous Poisson
process Y on B with intensity ρ0 . Next we obtain XB as an independent
thinning of YB with retention probabilities p(ξ) = ρ(ξ)/ρ0 , ξ ∈ B, cf.
Proposition 3.7. The right plot in Figure 3.1 is generated in this way.
For d = 1, an alternative is the transformation method considered in
Remark 3.4 (this method does not require ρ to be bounded on B).
3.2.4 Densities for Poisson processes

If X1 and X2 are two point processes defined on the same space S, then
X1 is absolutely continuous with respect to X2 (or more precisely the

distribution of X1 is absolutely continuous with respect to the distribu-
tion of X2 ) if and only if P (X2 ∈ F ) = 0 implies that P (X1 ∈ F ) = 0
for F ⊆ Nlf . Equivalently, by the Radon-Nikodym theorem (see e.g. p.
422 in Billingsley 1995), there exists a function f : Nlf → [0, ∞] so that

P (X1 ∈ F ) = E 1[X2 ∈ F ]f (X2 ) , F ⊆ Nlf .
We call f a density for X1 with respect to X2 .
The following proposition shows that Poisson processes are not always
absolutely continuous with respect to each other; but they are always
absolutely continuous with respect to the standard Poisson process if we
let S be bounded.
Proposition 3.8
(i) For any numbers ρ1 > 0 and ρ2 > 0, Poisson(Rd , ρ1 ) is absolutely
continuous with respect to Poisson(Rd , ρ2 ) if and only if ρ1 = ρ2 .

(ii) For i = 1, 2, suppose that ρi : S → [0, ∞) so that µi (S) = S ρi (ξ)dξ
is finite, and that ρ2 (ξ) > 0 whenever ρ1 (ξ) > 0. Then Poisson(S, ρ1 )
is absolutely continuous with respect to Poisson(S, ρ2 ), with density

f (x) = exp(µ2 (S) − µ1 (S)) ρ1 (ξ)/ρ2 (ξ) (3.11)
ξ∈x
for finite point configurations x ⊂ S (taking 0/0 = 0).
Proof. (i) The “if” part is trivial. Assume that ρ1 = ρ2 . Consider dis-
i m⊆ R with |Ai | = 1, i = 1, 2, . . ., and let Fd = {x ∈
d
joint subsets A
Nlf : limm→∞ i=1 n(x ∩ Ai )/m = ρ1 }. Under both Poisson(R , ρ1 ) and
Poisson(Rd , ρ2 ), the N (Ai ) are i.i.d. but with means ρ1 and ρ2 , respec-
tively. Hence, by the strong law of large numbers, P (X1 ∈ F ) = 1 but
P (X2 ∈ F ) = 0.
(ii) IfY is Poisson(S, ρ2 ), we have to verify that P (X ∈ F ) = E 1[Y ∈
F ]f (Y ) for F ⊆ Nlf . This is straightforwardly derived from Proposi-
tion 3.1.
We consider densities of point processes in more detail in Chapter 6.
3.3 Marked Poisson pro cesses

Consider now a marked point process X = {(ξ, mξ ) : ξ ∈ Y } with
points in T and mark space M , cf. Section 2.2 or the general setting in
Section B.2.
Definition 3.7 Suppose that Y is Poisson(T, φ), where φ is a locally

integrable intensity function, and conditional on Y, the marks {mξ :

ξ ∈ Y } are mutually independent. Then X is a marked Poisson process.
If the marks are identically distributed with a common distribution Q,
then Q is called the mark distribution.
The marks can e.g. be integers, real numbers, geometric objects, or

even themselves point processes. If M = {1, . . . , k}, then X is called
a multitype Poisson process. If M = {K ⊆ S : K compact}, then
the so-called Boolean model is obtained (more details can be found in
Molchanov 1997).
A simulation of a Boolean model is shown in Figure 3.2 (left). For com-
parison with the spruces data in Example 1.4 we use T = [0, 56] × [0, 38]
and sample the marks from the empirical distribution of the disc radii
in Figure 1.4 (see also Figure 4.13 in Section 4.5). The large number
of overlapping discs in the simulation indicates that a Boolean model is
not appropriate for the spruces data. In Examples 9.1 and 9.2 in Sec-
tions 9.1.4 we consider more appropriate Markov point process models.
. . . . .
.
. .
. . . . .
. . .. .
. .. .
. .
. . .
. . .. .
. . . . . .
. . . . . .
.
. . . . . .
. .
. . . ..
. . .
. . . .
. . . . . .. .
.. . .. .
.
.. .. . .
. . .
. . . . .
.. .
. . . .
.. . .. . . .
. .
. . . .
. . .. .
. . .. .
. . . . .
. . .
.
Figure 3.2 Left: simulation of a homogeneous Boolean model conditional on

134 points and with disc radii sampled from the empirical distribution of the
disc radii in Figure 1.4. Right: simulation of a bivariate Poisson process where
circles are given by random displacements of the crosses using a bivariate
standard normal distribution. In both plots the simulation window is [0, 56] ×
[0, 38].
In the rest of this section we consider the case M ⊆ Rp , p ≥ 1.
Proposition 3.9 Let X be a marked Poisson process with M ⊆ Rp

and where, conditional on Y , each mark mξ has a discrete or continuous
density pξ which does not depend on Y \ ξ. Let ρ(ξ, m) = φ(ξ)pξ (m).
Then
(i) X ∼ Poisson(T × M, ρ);

(ii) if the density on M defined by κ(m) = T
ρ(ξ, m)dξ is locally in-

tegrable, then {mξ : ξ ∈ Y } ∼ Poisson(M, κ) (where we refer to
Remark 3.1 for the discrete case of M ).
Proof. (i) follows directly from Proposition 3.1.

(ii) Let B be a bounded Borel set in M . Then

P ({mξ : ξ ∈ Y }∩B = ∅) = 1−P (X ∩(T ×B) = ∅) = exp(− κ(m)dm)
B
where the last equality follows from (i). The result now follows from
Theorem 3.1.
Note that not all Poisson(T × M, ρ)-processes are marked Poisson

processes,
since Definition 3.7 requires local
integrability of φ(ξ) given by
M ρ(ξ, m)dm in the continuous case or by ξ∈M ρ(ξ, m) in the discrete
case. Likewise, it is easy to construct marked Poisson processes which
are not Poisson processes, since the conditional distribution of mξ for a
marked Poisson process may depend on Y \ ξ.
3.3.1 Random independent displacements of the points in a Poisson

process
Suppose that Y ∗ = {ξ + mξ : ξ ∈ Y } is obtained by random independent
displacements of the points in Y ∼ Poisson(Rd , ρ) where conditional on
Y , the mξ are independent and each distributed according to a density
pξ on Rd which does not depend on Y \ ξ. Letting X ∗ be the marked
Poisson point process with marked points (ξ, ξ + mξ ), it follows from (ii)
in Proposition 3.9 that Y ∗ is a Poisson process with intensity function

∗
ρ (η) = ρ(ξ)pξ (η − ξ)dξ
provided ρ∗ is locally integrable. If ρ is constant and pξ does not depend

on ξ, then Y ∗ becomes a stationary Poisson process with intensity ρ∗ =
ρ.
Figure 3.2 (right) shows a simulation of a bivariate Poisson process,
where one type of points (the circles) are random displacements of the
other type of points (the crosses), which follow a Poisson([0, 56] × [0, 38],
100)-process. The displacements are given by independent bivariate stan-
dard normal distributions.
3.3.2 Multivariate Poisson processes and random labelling

By a multivariate Poisson process (or a bivariate Poisson process if k =
2) is usually meant that each Xi is a stationary Poisson process on Rd
with intensity 0 < ρi < ∞ for i = 1, . . . , k, see e.g. Diggle (1983). An

example of a bivariate Poisson process with positive correlation between
the components is (Y, Y ∗ ) where Y is a stationary Poisson process and
Y ∗ is obtained by random independent displacements of the points in Y .
A construction of a bivariate Poisson process with negative correlation
between the components is given in Brown, Silverman & Milne (1981).
Consider now a multitype process X with M = {1, . . . , k}. This can be
identified with a multivariate point process (X1 , . . . , Xk ), cf. Section 2.2.
We have equivalence between the following two properties: (i) P (mξ =
i|Y = y) = pξ (i) depends only on ξ for realisations y of Y and ξ ∈
y; (ii) (X1 , . . . , Xk ) is a multivariate Poisson process with independent
components Xi ∼ Poisson(T, ρi ) where ρi (ξ) = φ(ξ)pξ (i), i = 1, . . . , k.
That (i) implies (ii) follows by induction using Proposition 3.7, since if
e.g. k = 2, we can think of X1 = {ξ ∈ Y : mξ = 1} as an independent
thinning of Y with retention probability pξ (1), ξ ∈ Y . That (ii) implies
(i) is a consequence of the well-known result that we obtain a multinomial
distribution when conditioning on the sum of k independent Poisson
variables.
A common hypothesis for marked point processes {(ξ, mξ ) : ξ ∈ Y } is
that of random labelling which means that conditional on Y , the marks
mξ are mutually independent and the distribution of mξ does not depend
on Y . For a multitype Poisson process, for example, random labelling
means (i) above with pξ (i) = p(i) not depending on the location ξ.

CHAPTER 4
Summary statistics
This chapter surveys the most commonly used summary statistics for
point processes and multitype point processes in the homogeneous as
well as the inhomogeneous case. Exploratory analysis for spatial point
patterns and the validation of fitted models are often based on nonpara-
metric estimates of various summary statistics. For example, the initial
step of a point process analysis often consists of looking for discrepan-
cies with a Poisson model using nonparametric estimates of summary
statistics.
Section 4.1 deals with characteristics describing the first and second
order properties of a point process, particularly intensity and pair corre-
lation functions and second order reduced moment measures. Section 4.2
focuses on summary statistics for second order properties (such as the
so-called K and L-functions, and including the anisotropic case) and
summary statistics based on interpoint distances (such as the so-called
F , G, and J-functions). Section 4.3 concerns nonparametric estimation
of summary statistics and contains some examples of applications; many
other examples are given in the other chapters of this book. Sections 4.4–
4.5 concern summary statistics for multivariate and marked point pro-
cesses. The more technical part related to this chapter is deferred to
Appendix C.
Further material on summary statistics can be found in Stoyan &
Stoyan (1994, 2000), Baddeley & Gill (1997), Ohser & Mücklich (2000),
and the references therein. The application of many of the summary
statistics in this chapter requires some sort of stationarity assumption.
We therefore consider point processes (or marked point processes) with
points in Rd . However, in some cases we can replace Rd with a proper
subset of Rd or an arbitrary space S, cf. Appendix C.
4.1 First and second order prop erties of a p oint pro cess
Throughout this section X denotes a point process on S = Rd .

4.1.1 Basic definitions and results
The first and second order properties of the random count variables
N (B) for B ⊆ S are described by the so-called intensity measure and
second order factorial moment measure defined as follows.
Definition 4.1 The intensity measure µ on Rd is given by

µ(B) = EN (B), B ⊆ Rd ,
and the second order factorial moment measure α(2) on Rd × Rd by
=

α(2) (C) = E 1[(ξ, η) ∈ C], C ⊆ Rd × Rd .
ξ,η∈X
In Appendix C we consider in more detail moment measures of any

order on a general space S. Note that
E N (B1 )N (B2 ) = α(2) (B1 × B2 ) + µ(B1 ∩ B2 ), B1 , B2 ⊆ Rd , (4.1)
so α(2) and µ determine the second order moments of the random vari-
ables N (B), B ⊆ Rd .
Definition 4.2 If the intensity measure µ can be written as

µ(B) = ρ(ξ)dξ, B ⊆ Rd ,
B
where ρ is a nonnegative function, then ρ is called the intensity function.
If ρ is constant, then X is said to be homogeneous or first order stationary
with intensity ρ; otherwise X is said to be inhomogeneous.
Definition 4.2 is in agreement with Definition 3.3. Heuristically, ρ(ξ)dξ
is the probability for the occurrence of a point in an infinitesimally small
ball with centre ξ and volume dξ. For a homogeneous point process, ρ
is the mean number of points per unit volume.
Definition 4.3 If the second order factorial moment measure α(2) can
be written as

α(2) (C) = 1[(ξ, η) ∈ C]ρ(2) (ξ, η)dξdη, C ⊆ Rd × Rd ,
where ρ(2) is a nonnegative function, then ρ(2) is called the second order
product density.
Intuitively, ρ(2) (ξ, η)dξdη is the probability for observing a pair of

points from X occuring jointly in each of two infinitesimally small balls
with centres ξ, η and volumes dξ, dη. In order to study whether a point

process deviates from the Poisson process, it is useful to normalise the
second order product density ρ(2) (ξ, η) by dividing with ρ(ξ)ρ(η).
Definition 4.4 If both ρ and ρ(2) exist, the pair correlation function is
defined by
ρ(2) (ξ, η)
g(ξ, η) =
ρ(ξ)ρ(η)
where we take a/0 = 0 for a ≥ 0.
The g-function is widely used in astronomy and astrophysics (Peebles

1974, Besag 1977b, Kerscher 2000). For a Poisson process we imme-
diately obtain from the extended Slivnyak-Mecke Theorem 3.3 (Sec-
tion 3.2) that we can take ρ(2) (ξ, η) = ρ(ξ)ρ(η) so that g(ξ, η) = 1.
If for example g(ξ, η) > 1, this indicates that pair of points are more
likely to occur jointly at the locations ξ, η than for a Poisson process
with the same intensity function as X.
If X is stationary, g becomes translation invariant, i.e. g(ξ, η) = g(ξ −
η). † Note that there exist cases where g is translation invariant but ρ is
inhomogeneous, see e.g. Section 5.6.
Proposition 4.1 Suppose that X has intensity function ρ and second

order product density ρ(2) . Then for functions h1 : Rd → [0, ∞) and
h2 : Rd × Rd → [0, ∞),

E h1 (ξ) = h1 (ξ)ρ(ξ)dξ (4.2)
ξ∈X
and
=

E h2 (ξ, η) = h2 (ξ, η)ρ(2) (ξ, η)dξdη. (4.3)
ξ,η∈S
Proof. By the standard proof these equalities follow directly from Defi-
nitions 4.1–4.3.
A more general version of these results is given by (C.1) in Section C.1.1.
Proposition 4.2 Suppose that X has intensity function ρ and sec-

ond order product density ρ(2) , and that Xthin is an independent thin-
ning of X with retention probabilities p(ξ), ξ ∈ Rd . Then the inten-
sity function and second order product density of Xthin are given by
† Strictly speaking this holds for all ξ, η ∈ Rd \ A where A ⊂ Rd is of zero volume.

Note that we abuse notation and denote by g also the function which describes
how the pair correlation depends on differences.

(2)
ρthin (ξ) = p(ξ)ρ(ξ) and ρthin (ξ, η) = p(ξ)p(η)ρ(2) (ξ, η), and the pair cor-
relation function is invariant under independent thinning: g = gthin .
Proof. Using a notation as in Definition 3.6, recall that R(ξ) ∼ Uniform
([0, 1]), ξ ∈ Rd , are mutually independent and independent of X. So for
B ⊆ Rd ,

E n(Xthin ∩ B) = E E 1[ξ ∈ B, R(ξ) ≤ p(ξ)] X
ξ∈X

=E p(ξ)1[ξ ∈ B] = p(ξ)ρ(ξ)dξ
ξ∈X B
where the last equality follows from (4.2). Hence ρthin (ξ) = p(ξ)ρ(ξ). It
(2)
follows in the same way that ρthin (ξ, η) = p(ξ)p(η)ρ(2) (ξ, η), and so
(2)
gthin(ξ, η) = ρthin (ξ, η)/(ρthin (ξ)ρthin (η)) = ρ(2) (ξ, η)/(ρ(ξ)ρ(η))
= g(ξ, η).
4.1.2 The second order reduced moment measure

The second order factorial moment measure can sometimes be expressed
in terms of the intensity function and the following measure K on Rd .
Definition 4.5 Suppose that X has intensity function ρ and that the
measure
=

1 1[ξ ∈ A, η − ξ ∈ B]
K(B) = E , B ⊆ Rd , (4.4)
|A| ρ(ξ)ρ(η)
ξ,η∈X
does not depend on the choice of A ⊆ Rd with 0 < |A| < ∞, where we
take a/0 = 0 for a ≥ 0. Then X is said to be second order intensity
reweighted stationary and K is called the second order reduced moment
measure.
Stationarity of X implies second order intensity reweighted stationar-
ity.† If the pair correlation function exists and is invariant under trans-
† This follows by some elementary measure theoretical properties: In the stationary
case where ρ is constant,
=

ν(A) = |A|ρ2 K(B) = E 1[ξ ∈ A, η − ξ ∈ B]
ξ,η∈X
is seen to be a translation invariant measure for A ⊆ Rd when B is fixed, and so

ν is proportional to Lebesgue measure on Rd .

lations, then we have second order intensity reweighted stationarity and

K(B) = g(ξ)dξ, B ⊆ Rd . (4.5)
B
This follows by applying (4.3) to the right hand side in (4.4). Note that
K is in general not translation invariant (even if X is stationary).
When the intensity ρ is constant we have that

α (B1 × B2 ) = ρ
(2) 2
K(B2 − ξ)dξ.
B1
In the stationary case ρK(B) can be interpreted as the conditional expec-

tation of the number of further points in B given that X has a point at
the origin. This interpretation is given a precise meaning in Section C.2.3.
The measure K is used in Sections 4.2.1–4.2.2 for constructing vari-
ous summary statistics. A useful property, which holds for K and hence
the related summary statistics, is that K is invariant under independent
thinning:
Proposition 4.3 If Xthin is an independent thinning of a second or-

der reweighted stationary point process X, then Xthin is second order
reweighted stationary, and Xthin and X have the same K-measure.
Proof. This follows along similar lines as in the proof of Proposition 4.2.
4.2 Summary statistics

In this section we consider various summary statistics for a point process
X on Rd : the so-called K, L, g-functions in Section 4.2.1, directional K-
functions in Section 4.2.2, and F, G, J-functions in Section 4.2.3.
4.2.1 Second order summary statistics

In applications we consider estimates of K(B) for a class of test sets B
such as balls. Thereby second order summary statistics are obtained as
described below.
Definition 4.6 The K and L-functions for a second order reweighted

stationary point process are defined by
K(r) = K(b(0, r)), L(r) = (K(r)/ωd )1/d , r > 0.

This definition, which extends the definition of Ripley’s K-function
(Ripley 1976, 1977) for the stationary case to the case of second order in-
tensity reweighted stationarity, is due to Baddeley, Møller & Waagepeter-
sen (2000). In the stationary case, ρK(r) is the expected number of fur-
ther points within distance r from the origin given that X has a point
at the origin.
The K and L-functions are in one-to-one correspondence, and in ap-
plications the L-function is often used instead of the K-function. One
reason is that L is the identity for a Poisson process. In general, at least
for small values of r, L(r) − r > 0 indicates aggregation or clustering at
distances less than r, and L(r)− r < 0 regularity at distances less than r.
This may be due to certain latent processes (Chapter 5) or attraction or
repulsion between the points (Chapter 6). Moreover, for a homogeneous
Poisson process, the transformation K → L is variance stabilising when
K is estimated by nonparametric methods (Besag 1977b).
If K is invariant under rotations, then K is determined by K. This is
the case if X is isotropic, cf. (4.4), or if g(ξ, η) = g(ξ − η) is isotropic,†
cf. (4.5). If g is isotropic, then by (4.5),
r
K(r) = σd td−1 g(t)dt. (4.6)
0
This shows the close relationship between K and g. Many statisticians

seem more keen on using the L or K-function than g (see e.g. Ripley 1981
and Diggle 1983), possibly because it is simpler to estimate K than g,
cf. Section 4.3.2. However, since K is a cumulative function, it is usually
easier to interpret a plot of g than a plot of K. We shall therefore also
consider estimates of g when it is isotropic. In general, at least for small
values of r, g(r) > 1 indicates aggregation or clustering at distances r,
and g(r) < 1 regularity at such distances.
It should be noticed that very different point process models can share
the same K-function, see Baddeley & Silverman (1984) and Baddeley,
Møller & Waagepetersen (2000). In Section 5.4 we show that different
shot noise Cox processes can have the same pair correlation function.
4.2.2 Directional K-functions

Test sets B other than balls may be considered for K(B). For example,
in the planar case d = 2, in order to investigate for a possible anisotropy,
directional K-functions can be constructed as follows (see also Stoyan &
Stoyan 1994 and Brix & Møller 2001).
† Here and in the following we abuse notation and denote by g also the function
which describes how the pair correlation depends on interpoint distances.

For −π/2 ≤ ϕ < π/2, ϕ < ψ ≤ ϕ + π, and r > 0, define
K(ϕ, ψ, r) = K(B(ϕ, ψ, r)) (4.7)
where
B(ϕ, ψ, r) = {t(cos v, sin v) : 0 ≤ t ≤ r, φ ≤ v ≤ ψ or φ + π ≤ v ≤ ψ + π}
is the union of the two sectors of b(0, r) with the angles of the first sector
between ϕ and ψ, and the angles of the second sector between ϕ + π and
ψ + π, see Figure 4.1. If X is isotropic, then K(ϕ, ψ, r) = K(r)(ψ − ϕ)/π
for all (ϕ, ψ).
Figure 4.1 B(−π/6, π/6, r) (light grey).
4.2.3 Summary statistics based on interpoint distances

We now consider three summary statistics based on interpoint distances.
Definition 4.7 Assume that X is stationary. The empty space function

F is the distribution function of the distance from the origin (or another
fixed point in Rd ) to the nearest point in X, i.e.
F (r) = P (X ∩ b(0, r) = ∅), r > 0. (4.8)
The nearest-neighbour function G is
1
G(r) = E 1[(X \ ξ) ∩ b(ξ, r) = ∅], r > 0, (4.9)
ρ|A|
ξ∈X∩A
for an arbitrary set A ⊂ Rd with 0 < |A| < ∞. Finally, the J-function
is defined by
J(r) = (1 − G(r))/(1 − F (r)) for F (r) < 1. (4.10)
It is not obvious how to extend the definitions of F, G, and J to the

nonstationary case. The empty space function is also called the spherical
contact distribution function. By stationarity, (4.9) does not depend on

the choice of A. As indicated by its name, G can be interpreted as the
distribution function of the distance from a typical point in X to the
nearest-neighbour in X, see Section C.2.3. The J-function was suggested
by van Lieshout & Baddeley (1996).
For a stationary Poisson process on Rd with intensity ρ < ∞,
F (r) = G(r) = 1 − exp(−ρωd rd ) and J(r) = 1 for r > 0,
where the equality for G follows from the Slivnyak-Mecke Theorem 3.2
(Section 3.2). For other kind of models, closed form expressions of F, G, J
are rarely known. In general, at least for small values of r > 0, F (r) <
G(r) (or J(r) < 1) indicates aggregation or clustering, and F (r) > G(r)
(or J(r) > 1) regularity. Bedford & van den Berg (1997) show that J = 1
does not imply that X is a stationary Poisson process.
4.3 Nonparametric estimation

We now turn to nonparametric estimation of summary statistics. In this
section we let X be a spatial point process on Rd with intensity function
ρ; if X is stationary, ρ is assumed to be a constant with 0 < ρ < ∞.
Whenever needed we assume that the pair correlation function g or the
measure K exists. We confine ourself to the case where a single point pat-
tern XW = x is observed in a bounded window W ⊂ Rd with |W | > 0,
and discuss nonparametric estimation of ρ, K, K, L, g, F, G, and J. Ex-
tensions to replicated point patterns and to marked point processes are
sometimes obvious; see Diggle, Lange & Beněs (1991), Baddeley, Moy-
eed, Howard & Boyde (1993), Diggle, Mateu & Clough (2000), Schlather
(2001), and the references therein. Higher order summary statistics can
be introduced as well, but the corresponding nonparametric estimators
may be less stable if the number of points observed is not sufficiently
large; see Peebles & Groth (1975), Stoyan & Stoyan (1994), Møller,
Syversveen & Waagepetersen (1998), and Schladitz & Baddeley (2000).
4.3.1 Nonparametric estimation of intensity functions

In the homogeneous case, a natural unbiased estimate of the intensity is
ρ̂ = n(x)/|W |. (4.11)
This is in fact the maximum likelihood estimate if X is a homogeneous
Poisson process.
In the inhomogeneous case, a nonparametric kernel estimate of the
intensity function is

ρ̂b (ξ) = kb (ξ − η)/cW,b (η), ξ ∈ W (4.12)
η∈x

(Diggle 1985). Here kb is a kernel with band width b > 0, i.e. kb (ξ) =
k(ξ/b)/bd where k is a given density function, and

cW,b (η) = kb (ξ − η)dξ
W
is an edge correction factor. The estimate (4.12) is usually sensitive to

the choice of b, while the choice of k is less important. In our examples
where d = 2 we use a product kernel given by k(ξ) = e(ξ1 )e(ξ2 ) for
ξ = (ξ1 , ξ2 ) ∈ R2 , where
e(u) = (3/4)(1 − |u|)1[|u| ≤ 1], u ∈ R, (4.13)
is the Epanečnikov kernel.

Lemma 4.1 W
ρ̂b (ξ)dξ is an unbiased estimate of µ(W ).
Proof. We have that

E kb (ξ − η)/cW,b (η)dξ = E kb (ξ − η)/cW,b (η)dξ
W η∈X W η∈XW

W
= (kb (ξ − η)/cW,b (η))ρ(η)dηdξ = ρ(η)dη = µ(W )

W W W
using (4.2) in the second equality.
Example 4.1 (Intensity function for mucous cell data) Figure 4.2 shows
nonparametric estimates of the intensity function for the point pattern
in Example 1.3 when the information of point type is ignored. The band
widths used are b = 0.08, 0.12, 0.16, and 0.20. The choice b = 0.16 seems
to provide a suitable impression of the large scale variation in the point
pattern.
4.3.2 Nonparametric estimation of K and L

For nonparametric estimation of K, it is useful to establish the following
lemma where we write Wξ = {η + ξ : η ∈ W } for the translate of W by
ξ ∈ Rd .
Lemma 4.2 Suppose that |W ∩ Wξ | > 0 for all ξ ∈ B, and that X is

second order intensity reweighted stationary. Then
=
1[η − ξ ∈ B]
(4.14)
ρ(ξ)ρ(η)|W ∩ Wη−ξ |
ξ,η∈x
is an unbiased estimate of K(B).

Figure 4.2 Nonparametric estimates of the intensity function for the mucous
membrane cell data obtained with band widths b = 0.08 (upper left), 0.12 (upper
right), 0.16 (lower left), and 0.20 (lower right).
Proof. For simplicity assume that X has a translation invariant pair cor-
relation function g so that we can use (4.3) with ρ(2) (ξ, η) = ρ(η)ρ(ξ)g(η−
ξ). In the general case the proof is similar, using (C.2) instead of (4.3).
Letting h(ξ, η) = 1[η − ξ ∈ B]/(ρ(ξ)ρ(η)|W ∩ Wη−ξ |) in (4.3) we get
=
1[η − ξ ∈ B]
E
ρ(ξ)ρ(η)|W ∩ Wη−ξ |
ξ,η∈XW

1[ξ ∈ W, η ∈ W, η − ξ ∈ B]
= g(η − ξ)dηdξ
|W ∩ Wη−ξ |

1[ξ ∈ W ∩ W−η̃ , η̃ ∈ B]
= g(η̃)dξdη̃
|W ∩ Wη̃ |
= K(B)
using (4.5) and that |W ∩ Wη̃ | = |W ∩ W−η̃ | in the last equality.
The condition on B in Lemma 4.2 means that if e.g. B = b(0, r) and

W is rectangular, it is required that r be smaller than the smallest side in
W . The condition can be weakened: for instance, if the pair correlation

exists, it suffices to assume that
|{ξ ∈ B : |W ∩ Wξ | = 0}| = 0. (4.15)
We illustrate later in Example 4.3 what this means for the special design
of the weed plants.
Lemma 4.2 provides an unbiased estimate of K(B) provided ρ is known.
In practice ρ is not known, so ρ(ξ)ρ(η) in (4.14) must be replaced by an
The combined estimate
estimate ρ(ξ)ρ(η).
=
1[η − ξ ∈ B]
K̂(B) = (4.16)

ξ,η∈x ρ(ξ)ρ(η)|W ∩ Wη−ξ |
is then biased. The estimate of L(r) obtained from transforming that of

K(r) is in general biased as well.
In fact unbiasedness is usually unobtainable for many estimators in
spatial statistics, but instead they are often ratio-unbiased, i.e. of the
form θ̂ = Y /Z where θ = EY /EZ. For example, in the homogeneous
= ρ2 is unbiased, then (4.16) is ratio-unbiased. Stoyan &
case, if ρ(ξ)ρ(η)
Stoyan (2000) discuss various possibilities for the homogeneous case: one
possibility is to transform the estimate in (4.11) to obtain n(x)2 /|W |2
as an estimate of ρ2 ; an alternative is
ρ2 = n(x)(n(x) − 1)/|W |2 (4.17)
which is unbiased for a Poisson process. For the inhomogeneous case,
= ρ̄b (ξ)ρ̄b (η) where
Baddeley et al. (2000) propose to use ρ(ξ)ρ(η)

ρ̄b (ξ) = κb (ξ − η)/cW,b (η), ξ ∈ W, (4.18)
η∈x\ξ
is a slight modification of (4.12). Baddeley et al. (2000) show that for an

inhomogeneous Poisson processes, ρ̄b (ξ) is less biased than ρ̂b (ξ) when
ξ ∈ x is a data point. They argue that this is also the case for a point
process with pair correlation function g ≥ 1, while the picture is less
clear if g ≤ 1.
4.3.3 Edge correction

The weight 1/|W ∩Wη−ξ | in (4.14) and (4.16) is an edge correction factor
which we use because of its simplicity and general applicability; see also
the discussion in Stoyan & Stoyan (1994). However, numerous other edge
correction factors have been suggested in the literature, see e.g. Stoyan
& Stoyan (1994, 2000) and Ohser & Mücklich (2000).
A simpler alternative is based on minus sampling: For r > 0, let
W r = {ξ ∈ W : b(ξ, r) ⊆ W }

denote the set of points in W with a distance to the boundary of W
which is greater than r. Then
=
1[η − ξ ∈ B]
ρ(ξ)ρ(η)
ξ∈x,η∈x∩Wr
is an unbiased estimate of K(B) (which of course becomes biased when

we replace ρ(ξ)ρ(η) by an estimate). This is called the simple border
correction estimate or reduced-sample estimate of K(B). There is a loss
of information in this estimate, since some pairs of points are excluded. In
contrast the sum in (4.16) includes all observed pairs of points. However,
given sufficient data, one may prefer the reduced-sample estimate if in
(4.16) very large weights 1/|W ∩Wη−ξ | are given to point pairs {ξ, η} ⊆ x
with η − ξ ∈ B.
4.3.4 Envelopes for summary statistics

A plot of a nonparametric estimate of a summary statistic like L(r)
may be supplied by a confidence interval for each value of r. In the
following we consider the L-function, but the techniques apply on any
of the summary statistics considered in this chapter.
Consider a simple hypothesis H. Confidence intervals and other distri-
butional characteristics associated with the nonparametric estimate L̂(r)
can be obtained by a bootstrap using simulation under H. For a given
distance r > 0, let T0 (r) = T (X, r) denote L̂(r) obtained from the point
process X observed within the window W . Let T1 (r) = T (X1 , r), . . . ,
Tn (r) = T (Xn , r) be obtained from i.i.d. simulations X1 , . . . , Xn under
H. From the empirical distribution of T1 (r), . . . , Tn (r) we can estimate
any quantile for the distribution of T0 (r) under H, and we can do this
with any desired precision if n is large enough. Notice that although
T1 (r), . . . , Tn (r) are i.i.d., the random vectors (T1 (r), . . . , Tn (r)) consid-
ered for different values of r > 0 are dependent. So some caution should
be taken when we compare the results for different values of r.
If the computation of Ti (r), i = 1, . . . , n, is time consuming (this is
e.g. the case for the weed data in Example 1.5), the following envelopes
may be used where n is small (e.g. n = 39 as exemplified below). Let
Tmin (r) = min{T1 (r), . . . , Tn (r)} and Tmax (r) = max{T1 (r), . . . , Tn (r)}.
(4.19)
Under H,
P (T0 (r) < Tmin(r)) = P (T0 (r) > Tmax (r)) ≤ 1/(n + 1) (4.20)
with equality if T0 (r), T1 (r), . . . , Tn (r) are almost surely different. The
bounds Tmin(r) and Tmax (r) are called the 100/(n + 1)%-lower and the

Another random document with
no related content on Scribd:
4. BETTY GORDON AT BOARDING SCHOOL
or The Treasure of Indian Chasm
Seeking the treasure of Indian Chasm makes an exceedingly
interesting incident.
5. BETTY GORDON AT MOUNTAIN CAMP

or The Mystery of Ida Bellethorne
At Mountain Camp Betty found herself in the midst of a mystery
involving a girl whom she had previously met in Washington.
6. BETTY GORDON AT OCEAN PARK

or School Chums on the Boardwalk
A glorious outing that Betty and her chums never forgot.
7. BETTY GORDON AND HER SCHOOL CHUMS

or Bringing the Rebels to Terms
Rebellious students, disliked teachers and mysterious robberies
make a fascinating story.
THE RUTH FIELDING SERIES
By ALICE B. EMERSON
12mo. Illustrated. Price per volume, 65 cents, postpaid
Ruth Fielding will live in juvenile Fiction.
RUTH FIELDING OF THE RED MILL

or Jasper Parloe’s Secret
RUTH FIELDING AT BRIARWOOD HALL

or Solving the Campus Mystery
RUTH FIELDING AT SNOW CAMP

or Lost in the Backwoods
RUTH FIELDING AT LIGHTHOUSE POINT

or Nita, the Girl Castaway
RUTH FIELDING AT SILVER RANCH

or Schoolgirls Among the Cowboys
RUTH FIELDING ON CLIFF ISLAND

or The Old Hunter’s Treasure Box
RUTH FIELDING AT SUNRISE FARM
or What Became of the Raby Orphans
RUTH FIELDING AND THE GYPSIES

or The Missing Pearl Necklace
RUTH FIELDING IN MOVING PICTURES

or Helping the Dormitory Fund
RUTH FIELDING DOWN IN DIXIE

or Great Days in the Land of Cotton
RUTH FIELDING AT COLLEGE

or The Missing Examination Papers
RUTH FIELDING IN THE SADDLE

or College Girls in the Land of Gold
RUTH FIELDING IN THE RED CROSS

or Doing Her Bit for Uncle Sam
RUTH FIELDING AT THE WAR FRONT

or The Hunt for a Lost Soldier
RUTH FIELDING HOMEWARD BOUND

or A Red Cross Worker’s Ocean Perils
RUTH FIELDING DOWN EAST

or The Hermit of Beach Plum Point
RUTH FIELDING IN THE GREAT NORTHWEST

or The Indian Girl Star of the Movies
RUTH FIELDING ON THE ST. LAWRENCE

or The Queer Old Man of the Thousand Islands
RUTH FIELDING TREASURE HUNTING

or A Moving Picture that Became Real
CUPPLES & LEON COMPANY, Publishers New York
TRANSCRIBER’S NOTES
1. Silently corrected obvious typographical errors and
variations in spelling.
2. Retained archaic, non-standard, and uncertain spellings
as printed.
*** END OF THE PROJECT GUTENBERG EBOOK DOROTHY DALE TO
THE RESCUE ***
Updated editions will replace the previous one—the old editions

will be renamed.
Creating the works from print editions not protected by U.S.

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

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

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

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

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

Gutenberg:
1.E.1. The following sentence, with active links to, or other

immediate access to, the full Project Gutenberg™ License must
appear prominently whenever any copy of a Project
Gutenberg™ work (any work on which the phrase “Project
Gutenberg” appears, or with which the phrase “Project
Gutenberg” is associated) is accessed, displayed, performed,
viewed, copied or distributed:
This eBook is for the use of anyone anywhere in the United

States and most other parts of the world at no cost and
with almost no restrictions whatsoever. You may copy it,
give it away or re-use it under the terms of the Project
Gutenberg License included with this eBook or online at
www.gutenberg.org. If you are not located in the United
States, you will have to check the laws of the country
where you are located before using this eBook.
1.E.2. If an individual Project Gutenberg™ electronic work is

derived from texts not protected by U.S. copyright law (does not
contain a notice indicating that it is posted with permission of
the copyright holder), the work can be copied and distributed to
anyone in the United States without paying any fees or charges.
If you are redistributing or providing access to a work with the
phrase “Project Gutenberg” associated with or appearing on the
work, you must comply either with the requirements of
paragraphs 1.E.1 through 1.E.7 or obtain permission for the use
of the work and the Project Gutenberg™ trademark as set forth
in paragraphs 1.E.8 or 1.E.9.
1.E.3. If an individual Project Gutenberg™ electronic work is

posted with the permission of the copyright holder, your use and
distribution must comply with both paragraphs 1.E.1 through
1.E.7 and any additional terms imposed by the copyright holder.
Additional terms will be linked to the Project Gutenberg™
License for all works posted with the permission of the copyright
holder found at the beginning of this work.
1.E.4. Do not unlink or detach or remove the full Project

Gutenberg™ License terms from this work, or any files
containing a part of this work or any other work associated with
Project Gutenberg™.
1.E.5. Do not copy, display, perform, distribute or redistribute

this electronic work, or any part of this electronic work, without
prominently displaying the sentence set forth in paragraph 1.E.1
with active links or immediate access to the full terms of the
Project Gutenberg™ License.
1.E.6. You may convert to and distribute this work in any binary,
compressed, marked up, nonproprietary or proprietary form,
including any word processing or hypertext form. However, if
you provide access to or distribute copies of a Project
Gutenberg™ work in a format other than “Plain Vanilla ASCII” or
other format used in the official version posted on the official
Project Gutenberg™ website (www.gutenberg.org), you must,
at no additional cost, fee or expense to the user, provide a copy,
a means of exporting a copy, or a means of obtaining a copy
upon request, of the work in its original “Plain Vanilla ASCII” or
other form. Any alternate format must include the full Project
Gutenberg™ License as specified in paragraph 1.E.1.
1.E.7. Do not charge a fee for access to, viewing, displaying,

performing, copying or distributing any Project Gutenberg™
works unless you comply with paragraph 1.E.8 or 1.E.9.
1.E.8. You may charge a reasonable fee for copies of or

providing access to or distributing Project Gutenberg™
electronic works provided that:
• You pay a royalty fee of 20% of the gross profits you derive
from the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
about donations to the Project Gutenberg Literary Archive
Foundation.”
• You provide a full refund of any money paid by a user who

notifies you in writing (or by e-mail) within 30 days of receipt
that s/he does not agree to the terms of the full Project
Gutenberg™ License. You must require such a user to return or
destroy all copies of the works possessed in a physical medium
and discontinue all use of and all access to other copies of
Project Gutenberg™ works.
• You provide, in accordance with paragraph 1.F.3, a full refund of

any money paid for a work or a replacement copy, if a defect in
the electronic work is discovered and reported to you within 90
days of receipt of the work.
• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.
1.E.9. If you wish to charge a fee or distribute a Project

Gutenberg™ electronic work or group of works on different
terms than are set forth in this agreement, you must obtain
permission in writing from the Project Gutenberg Literary
Archive Foundation, the manager of the Project Gutenberg™
trademark. Contact the Foundation as set forth in Section 3
below.
1.F.
1.F.1. Project Gutenberg volunteers and employees expend

considerable effort to identify, do copyright research on,
transcribe and proofread works not protected by U.S. copyright
law in creating the Project Gutenberg™ collection. Despite these
efforts, Project Gutenberg™ electronic works, and the medium
on which they may be stored, may contain “Defects,” such as,
but not limited to, incomplete, inaccurate or corrupt data,
transcription errors, a copyright or other intellectual property
infringement, a defective or damaged disk or other medium, a
computer virus, or computer codes that damage or cannot be
read by your equipment.
1.F.2. LIMITED WARRANTY, DISCLAIMER OF DAMAGES - Except

for the “Right of Replacement or Refund” described in
paragraph 1.F.3, the Project Gutenberg Literary Archive
Foundation, the owner of the Project Gutenberg™ trademark,
and any other party distributing a Project Gutenberg™ electronic
work under this agreement, disclaim all liability to you for
damages, costs and expenses, including legal fees. YOU AGREE
THAT YOU HAVE NO REMEDIES FOR NEGLIGENCE, STRICT
LIABILITY, BREACH OF WARRANTY OR BREACH OF CONTRACT
EXCEPT THOSE PROVIDED IN PARAGRAPH 1.F.3. YOU AGREE
THAT THE FOUNDATION, THE TRADEMARK OWNER, AND ANY
DISTRIBUTOR UNDER THIS AGREEMENT WILL NOT BE LIABLE
TO YOU FOR ACTUAL, DIRECT, INDIRECT, CONSEQUENTIAL,
PUNITIVE OR INCIDENTAL DAMAGES EVEN IF YOU GIVE
NOTICE OF THE POSSIBILITY OF SUCH DAMAGE.
1.F.3. LIMITED RIGHT OF REPLACEMENT OR REFUND - If you

discover a defect in this electronic work within 90 days of
receiving it, you can receive a refund of the money (if any) you
paid for it by sending a written explanation to the person you
received the work from. If you received the work on a physical
medium, you must return the medium with your written
explanation. The person or entity that provided you with the
defective work may elect to provide a replacement copy in lieu
of a refund. If you received the work electronically, the person
or entity providing it to you may choose to give you a second
opportunity to receive the work electronically in lieu of a refund.
If the second copy is also defective, you may demand a refund
in writing without further opportunities to fix the problem.
1.F.4. Except for the limited right of replacement or refund set

forth in paragraph 1.F.3, this work is provided to you ‘AS-IS’,
WITH NO OTHER WARRANTIES OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.
1.F.5. Some states do not allow disclaimers of certain implied

warranties or the exclusion or limitation of certain types of
damages. If any disclaimer or limitation set forth in this
agreement violates the law of the state applicable to this
agreement, the agreement shall be interpreted to make the
maximum disclaimer or limitation permitted by the applicable
state law. The invalidity or unenforceability of any provision of
this agreement shall not void the remaining provisions.
1.F.6. INDEMNITY - You agree to indemnify and hold the

Foundation, the trademark owner, any agent or employee of the
Foundation, anyone providing copies of Project Gutenberg™
electronic works in accordance with this agreement, and any
volunteers associated with the production, promotion and
distribution of Project Gutenberg™ electronic works, harmless
from all liability, costs and expenses, including legal fees, that
arise directly or indirectly from any of the following which you
do or cause to occur: (a) distribution of this or any Project
Gutenberg™ work, (b) alteration, modification, or additions or
deletions to any Project Gutenberg™ work, and (c) any Defect
you cause.
Section 2. Information about the Mission

of Project Gutenberg™
Project Gutenberg™ is synonymous with the free distribution of
electronic works in formats readable by the widest variety of
computers including obsolete, old, middle-aged and new
computers. It exists because of the efforts of hundreds of
volunteers and donations from people in all walks of life.
Volunteers and financial support to provide volunteers with the

assistance they need are critical to reaching Project
Gutenberg™’s goals and ensuring that the Project Gutenberg™
collection will remain freely available for generations to come. In
2001, the Project Gutenberg Literary Archive Foundation was
created to provide a secure and permanent future for Project
Gutenberg™ and future generations. To learn more about the
Project Gutenberg Literary Archive Foundation and how your
efforts and donations can help, see Sections 3 and 4 and the
Foundation information page at www.gutenberg.org.
Section 3. Information about the Project

Gutenberg Literary Archive Foundation
The Project Gutenberg Literary Archive Foundation is a non-
profit 501(c)(3) educational corporation organized under the
laws of the state of Mississippi and granted tax exempt status
by the Internal Revenue Service. The Foundation’s EIN or
federal tax identification number is 64-6221541. Contributions
to the Project Gutenberg Literary Archive Foundation are tax
deductible to the full extent permitted by U.S. federal laws and
your state’s laws.
The Foundation’s business office is located at 809 North 1500

West, Salt Lake City, UT 84116, (801) 596-1887. Email contact
links and up to date contact information can be found at the
Foundation’s website and official page at
www.gutenberg.org/contact
Section 4. Information about Donations to
the Project Gutenberg Literary Archive
Foundation
Project Gutenberg™ depends upon and cannot survive without
widespread public support and donations to carry out its mission
of increasing the number of public domain and licensed works
that can be freely distributed in machine-readable form
accessible by the widest array of equipment including outdated
equipment. Many small donations ($1 to $5,000) are particularly
important to maintaining tax exempt status with the IRS.
The Foundation is committed to complying with the laws

regulating charities and charitable donations in all 50 states of
the United States. Compliance requirements are not uniform
and it takes a considerable effort, much paperwork and many
fees to meet and keep up with these requirements. We do not
solicit donations in locations where we have not received written
confirmation of compliance. To SEND DONATIONS or determine
the status of compliance for any particular state visit
www.gutenberg.org/donate.
While we cannot and do not solicit contributions from states

where we have not met the solicitation requirements, we know
of no prohibition against accepting unsolicited donations from
donors in such states who approach us with offers to donate.
International donations are gratefully accepted, but we cannot

make any statements concerning tax treatment of donations
received from outside the United States. U.S. laws alone swamp
our small staff.
Please check the Project Gutenberg web pages for current

donation methods and addresses. Donations are accepted in a
number of other ways including checks, online payments and
credit card donations. To donate, please visit:
www.gutenberg.org/donate.
Section 5. General Information About

Project Gutenberg™ electronic works
Professor Michael S. Hart was the originator of the Project
Gutenberg™ concept of a library of electronic works that could
be freely shared with anyone. For forty years, he produced and
distributed Project Gutenberg™ eBooks with only a loose
network of volunteer support.
Project Gutenberg™ eBooks are often created from several

printed editions, all of which are confirmed as not protected by
copyright in the U.S. unless a copyright notice is included. Thus,
we do not necessarily keep eBooks in compliance with any
particular paper edition.
Most people start at our website which has the main PG search
facility: www.gutenberg.org.
This website includes information about Project Gutenberg™,

including how to make donations to the Project Gutenberg
Literary Archive Foundation, how to help produce our new
eBooks, and how to subscribe to our email newsletter to hear
about new eBooks.

(FREE PDF Sample) Statistical Inference and Simulation For Spatial Point Processes 1st Edition Jesper Møller Ebooks

Uploaded by

Copyright:

Available Formats

(FREE PDF Sample) Statistical Inference and Simulation For Spatial Point Processes 1st Edition Jesper Møller Ebooks

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(FREE PDF Sample) Statistical Inference and Simulation For Spatial Point Processes 1st Edition Jesper Møller Ebooks

Uploaded by

Copyright:

Available Formats

Full download ebook at ebookgate.

Statistical inference and simulation for

Download more ebook from https://ebookgate.com

Statistical Inference for Fractional Diffusion

Statistical Inference 2nd Edition George Casella

Models for Probability and Statistical Inference Theory

Models for Probability and Statistical Inference Theory

Essentials Of Statistical Inference 1st Edition G. A.

Principles of Statistical Inference 1st Edition D. R.

Statistical Inference A Short Course 1st Edition

Instructor s Solutions Manual for Probability and

CHAPMAN & HALL/CRC

© 2004 by Chapman & Hall/CRC

Library of Congress Cataloging-in-Publication Data

Visit the CRC Press Web site at www.crcpress.com

© 2004 by Chapman & Hall/CRC

No claim to original U.S. Government works

1 Examples of spatial p oint patterns

2 Intro duction to p oint pro cesses

3 Poisson p oint pro cesses

© 2004 by Chapman & Hall/CRC

5 Cox pro cesses

6 Markov p oint pro cesses

© 2004 by Chapman & Hall/CRC

7 Metrop olis-Hastings algorithms

© 2004 by Chapman & Hall/CRC

9 Inference for Markov p oint pro cesses

10 Inference for Cox pro cesses

© 2004 by Chapman & Hall/CRC

11 Birth-death pro cesses and p erfect simulation

A History, bibliography, and software

B Measure theoretical details

© 2004 by Chapman & Hall/CRC

D Perfect simulation of SNCPs

E Simulation of Gaussian fields

F Nearest-neighb our Markov p oint pro cesses

G Results for spatial birth-death pro cesses

© 2004 by Chapman & Hall/CRC

© 2004 by Chapman & Hall/CRC

© 2004 by Chapman & Hall/CRC

We have been supported by the European Union’s TMR network “Sta-

© 2004 by Chapman & Hall/CRC

Examples of spatial point patterns

This chapter introduces some examples of spatial point patterns which

© 2004 by Chapman & Hall/CRC

Example 1.3 (Mucous membrane cells) The point pattern in Figure

Figure 1.1 Positions of hickory trees.

©2004 by Chapman & Hall/CRC

©2004 by Chapman & Hall/CRC

Figure 1.5 Sampling design for weed data.

©2004 by Chapman & Hall/CRC

This bivariate point pattern is a subset of a much larger dataset

©2004 by Chapman & Hall/CRC

©2004 by Chapman & Hall/CRC

Introduction to point processes

This chapter gives an informal introduction to point process and to

2.1 Point pro cesses on Rd

A spatial point process X is a random countable subset of a space S.

is the restriction of a point conﬁguration x to B (similarly XB is the