Theory of Spatial Statistics A Concise

Introduction 1st Edition M.N.M. Van

Lieshout
Visit to download the full and correct content document:
https://textbookfull.com/product/theory-of-spatial-statistics-a-concise-introduction-1st-
edition-m-n-m-van-lieshout/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

A Concise Introduction to Measure Theory Satish Shirali

https://textbookfull.com/product/a-concise-introduction-to-
measure-theory-satish-shirali/

Psychology A Concise Introduction 5th Edition Richard

A. Griggs

https://textbookfull.com/product/psychology-a-concise-
introduction-5th-edition-richard-a-griggs/

Mathematical Foundations of Time Series Analysis A

Concise Introduction 1st Edition Jan Beran

https://textbookfull.com/product/mathematical-foundations-of-
time-series-analysis-a-concise-introduction-1st-edition-jan-
beran/

A Concise Introduction to Pure Mathematics Fourth

Edition Liebeck

https://textbookfull.com/product/a-concise-introduction-to-pure-
mathematics-fourth-edition-liebeck/
Sociology A Concise South African Introduction Paul
Stewart

https://textbookfull.com/product/sociology-a-concise-south-
african-introduction-paul-stewart/

A Concise Introduction to Hypercomplex Fractals First

Edition Andrzej Katunin

https://textbookfull.com/product/a-concise-introduction-to-
hypercomplex-fractals-first-edition-andrzej-katunin/

Visual research a concise introduction to thinking

visually Crowder

https://textbookfull.com/product/visual-research-a-concise-
introduction-to-thinking-visually-crowder/

Lymphedema: A Concise Compendium of Theory and Practice

2nd Edition Byung-Boong Lee

https://textbookfull.com/product/lymphedema-a-concise-compendium-
of-theory-and-practice-2nd-edition-byung-boong-lee/

Rudiments of Music A Concise Guide to Music Theory

Third Edition Perry Goldstein

https://textbookfull.com/product/rudiments-of-music-a-concise-
guide-to-music-theory-third-edition-perry-goldstein/
Theory of Spatial
Statistics
A Concise Introduction
Theory of Spatial
Statistics
A Concise Introduction

M.N.M. van Lieshout

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742


c 2019 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-0-367-14642-9 (Hardback)

978-0-367-14639-9 (Paperback)

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

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

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

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

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

Library of Congress Cataloging-in-Publication Data

Names: van Lieshout, M. N. M., author.

Title: Theory of spatial statistics : a concise introduction / by M.N.M. van
Lieshout.
Description: Boca Raton, Florida : CRC Press, 2019. | Includes bibliograph-
ical references and index.
Identifiers: LCCN 2018052975| ISBN 9780367146429 (hardback : alk.
paper) | ISBN 9780367146399 (pbk. : alk. paper) | ISBN 9780429052866
(e-book)
Subjects: LCSH: Spatial analysis (Statistics)
Classification: LCC QA278.2 .V36 2019 | DDC 519.5/35- -dc23
LC record available at https://lccn.loc.gov/2018052975

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
To Catharina Johanna Schoenmakers.
Contents

Preface xi

Author xiii

Chapter 1  Introduction 1

1.1 GRIDDED DATA 1

1.2 AREAL UNIT DATA 3
1.3 MAPPED POINT PATTERN DATA 4
1.4 PLAN OF THE BOOK 6

Chapter 2  Random field modelling and interpolation 9

2.1 RANDOM FIELDS 9

2.2 GAUSSIAN RANDOM FIELDS 11
2.3 STATIONARITY CONCEPTS 13
2.4 CONSTRUCTION OF COVARIANCE FUNCTIONS 16
2.5 PROOF OF BOCHNER’S THEOREM 20
2.6 THE SEMI-VARIOGRAM 23
2.7 SIMPLE KRIGING 26
2.8 BAYES ESTIMATOR 28
2.9 ORDINARY KRIGING 29
2.10 UNIVERSAL KRIGING 32
2.11 WORKED EXAMPLES WITH R 34
2.12 EXERCISES 42
2.13 POINTERS TO THE LITERATURE 46

vii
viii  Contents

Chapter 3  Models and inference for areal unit data 49

3.1 DISCRETE RANDOM FIELDS 49

3.2 GAUSSIAN AUTOREGRESSION MODELS 52
3.3 GIBBS STATES 55
3.4 MARKOV RANDOM FIELDS 59
3.5 INFERENCE FOR AREAL UNIT MODELS 62
3.6 MARKOV CHAIN MONTE CARLO SIMULATION 67
3.7 HIERARCHICAL MODELLING 70
3.7.1 Image segmentation 70
3.7.2 Disease mapping 73
3.7.3 Synthesis 75
3.8 WORKED EXAMPLES WITH R 76
3.9 EXERCISES 84
3.10 POINTERS TO THE LITERATURE 88

Chapter 4  Spatial point processes 93

4.1 POINT PROCESSES ON EUCLIDEAN SPACES 93

4.2 THE POISSON PROCESS 96
4.3 MOMENT MEASURES 98
4.4 STATIONARITY CONCEPTS AND PRODUCT
DENSITIES 100
4.5 FINITE POINT PROCESSES 104
4.6 THE PAPANGELOU CONDITIONAL INTENSITY 108
4.7 MARKOV POINT PROCESSES 110
4.8 LIKELIHOOD INFERENCE FOR POISSON
PROCESSES 112
4.9 INFERENCE FOR FINITE POINT PROCESSES 114
4.10 COX PROCESSES 117
4.10.1 Cluster processes 118
4.10.2 Log-Gaussian Cox processes 120
4.10.3 Minimum contrast estimation 121
 Contents  ix

4.11 HIERARCHICAL MODELLING 123

4.12 WORKED EXAMPLES WITH R 126
4.13 EXERCISES 134
4.14 POINTERS TO THE LITERATURE 137

Appendix: Solutions to theoretical exercises 143

Index 167
Preface

Today, much information reaches us in graphical form. From a math-

ematical point of view, such data may be divided into various classes,
each having its own salient characteristics. For instance, in classical geo-
statistics, some spatially varying variable is observed at a given number
of fixed locations and one is interested in its value at locations where
it was not observed. One might think of the prediction of ore content
in the soil based on measurements at some conveniently placed bore-
holes or the construction of air pollution maps based on gauge data.
In other cases, due to technical constraints or for privacy reasons, data
is collected in aggregated form as region counts or as a discrete image.
Typical examples include satellite imagery, tomographic scans, disease
maps or yields in agricultural field trials. In this case, the objective is
often spatial smoothing or sharpening rather than prediction. Finally,
data may consist of a set of objects or phenomena tied to random spa-
tial locations and the prime interest is in the geometrical arrangement
of the set, for instance in the study of earthquakes or of cellular patterns
seen under a microscope.
The statistical analysis of spatial data merits treatment as a separate
topic, as it is different from ‘classical’ statistical data in a number of
aspects. Typically, only a single observation is available, so that artificial
replication in the form of an appropriate stationarity assumption is called
for. Also the size of images, or the number of objects in a spatial pattern,
is typically large and, moreover, there may be interactions at various
scales. Hence a conditional or hierarchical specification is useful, often
in combination with Monte Carlo methods.
This book will describe the mathematical foundations for each of
the data classes mentioned above, present some models and discuss sta-
tistical inference. Each chapter first presents the theory which is then
applied to illustrative examples using an open source R-package, lists
some exercises and concludes with pointers to the literature. The pre-
requisites consist of maturity in probability and statistics at the level
expected of a graduate student in mathematics, engineering or statistics.

xi
xii  Preface

Indeed, the contents grew out of lectures in the Dutch graduate school
‘Mastermath’ and are suitable for a semester long introduction. Those
wishing to learn more are referred to the excellent monographs by Cressie
(Wiley, 2015), by Banerjee, Carlin and Gelfand (CRC, 2004) and by
Gaetan and Guyon (Springer, 2010) or to the exhaustive Handbook of
Spatial Statistics (CRC, 2010).
In closing, I would like to express my gratitude to the students
who attended my ‘Mastermath’ courses for useful feedback, to the staff
at Taylor and Francis, especially to Rob Calver, for their support, to
three anonymous reviewers for constructive suggestions and to Christoph
Hofer–Temmel for a careful reading of the manuscript.

Marie-Colette van Lieshout

Amsterdam, September 2018
Author

M.N.M. van Lieshout is a senior researcher at the Centre for Mathe-

matics and Computer Science (CWI) in Amsterdam, The Netherlands,
and holds a chair in spatial stochastics at the University of Twente.

xiii
 CHAPTER 1

Introduction

The topic of these lecture notes is modelling and inference for spatial
data. Such data, by definition, involve measurements at some spatial
locations, but can take many forms depending on the stochastic mech-
anism that generated the data, on the type of measurement and on the
choice of the spatial locations.
Ideally, the feature of interest is measured at every location in some
appropriate region, usually a bounded subset of the plane. From a math-
ematical point of view, such a situation can be described by a random
field indexed by the region. In practice, however, it is not possible to
consider infinitely many locations. Additionally, there may be physical,
administrative, social or economic reasons for limiting the number of
sampling locations or for storing measurements in aggregated form over
areal units. The locations may even be random, so that, in mathematical
terms, they constitute a point process.
In the next three sections, we will present some typical examples
to motivate the more mathematical treatment in subsequent chapters.
Suggestions for statistical inference will also be given, but note that these
should be taken as an indication. Indeed, any pertinent analysis should
take into account the data collection process, the specific context and the
scientific question or goal that prompted data collection in the first place.

1.1 GRIDDED DATA

Figure 1.1 shows 208 coal ash core samples collected on a grid in the
Robena Mine in Greene County, Pennsylvania. The diameters of the
discs are proportional to the percentage of coal ash at the sampled loca-
tions. The data can be found in a report by Gomez and Hazen [1] and

1
2  Introduction

7
8.96
9.785
10.567
17.61

Percentage of coal ash sampled on a grid in the Robena Mine

Figure 1.1
in Greene County, Pennsylvania.

were prepared for R by E. Pebesma using a digital version at a website

maintained by D. Zimmerman.
A mining engineer might be interested in knowing basic summary
statistics, including the first and second moments of the sample. Next,
with model building in mind, he could ask himself whether the data
could have come from a normal distribution, possibly after discarding
some outliers, and if not, whether they are multi-modal or skewed. Such
questions could be addressed by elementary tools including histograms,
quantiles, boxplots and Q-Q plots.
On a higher conceptual level, the mining company could also be
interested in local outliers, measured percentages that are markedly dif-
ferent from those around them, or in trends that could indicate good
places to concentrate future mining efforts. Indications of these can be
found by applying the elementary statistics across rows or columns. For
instance, consideration of the mean across columns suggests that there
is a decreasing trend in the percentage of coal ash from left to right. It
 Areal unit data  3

would be of interest to quantify how strongly correlated measurements

at adjacent sampling locations are too.
Another striking feature of Figure 1.1 is that there are holes in the
sampling grid, for example in the seventh row from above and in the
sixth one from below. Therefore, when the mining engineer has found and
validated a reasonable model that accounts for both the global trends
and the local dependencies in the data, he could proceed to try and fill
in the gaps, in other words, to estimate the percentage of coal ash at
missing grid points based on the sampled percentages. Such a spatial
interpolation procedure is called kriging in honour of the South-African
statistician and mining engineer D.G. Krige, one of the pioneers of what
is now known as geostatistics.

1.2 AREAL UNIT DATA

The top-most panel of Figure 1.2 shows a graphical representation of
the total number of deaths from Sudden Infant Death Syndrome (SIDS)
in 1974 for each of the 100 counties in North Carolina. These data were
collected by the state’s public health statistics branch and analysed in
[2]. More precisely, the counts were binned in five colour-coded intervals,
where darker colours correspond to higher counts.
From the picture it is clear that the centroids of the counties do not
lie on a regular grid. The sizes and shapes of the counties vary and can
be quite irregular. Moreover, the recorded counts are not tied to a pre-
cise location but tallied up county-wise. This kind of accumulation over
administrative units is usual for privacy-sensitive data in, for instance,
the crime or public health domains.
A public health official could be interested in spatial patterns. Indeed,
the original research question in [2] was whether or not there are clusters
of counties with a high incidence of SIDS. However, death counts by
themselves are quite meaningless without an indication of the population
at risk. For this purpose, Symons, Grimson and Yuan [3] asked the North
Carolina public health statistics branch for the counts of live births in
each county during the same year 1974. These are shown in the lower
panel of Figure 1.2.
Presented with the two pictures, our public health official might look
for areas where the SIDS counts are higher than what would be expected
based on the number of live births in the area. Such areas would be
prime targets for an information campaign or a quest for factors specific
to those areas that could explain the outlier. For the data at hand, when
comparing counties at the north-east and the north-west with similar
4  Introduction

Figure 1.2 Numbers of cases of Sudden Infant Death Syndrome (top)

and live births (bottom) during 1974 accumulated per county in North
Carolina. The death counts are binned in five classes with breaks at
5, 10, 15 and 20, the live birth counts in six classes with breaks at 1000,
2000, 3000, 4000 and 5000. Darker colours correspond to higher counts.

birth numbers, it is clear that there is a higher SIDS rate in the north-
east. Note that there are also counties in the centre of the state with a
high number of births but a rather low SIDS incidence. Other outliers
can be identified using classic boxplots and quantile techniques on the
rates of SIDS compared to live births.
Such an analysis, however, ignores the fact that the county borders
are purely administrative and disease patterns are unlikely to follow.
Moreover, rates in counties with many births are likely to be more stable
than those with few. On a higher conceptual level, the public health
authority may therefore wish for a model that explicitly accounts for
large scale variations in expected rates and their associated variances as
well as for local dependencies between adjacent counties.

1.3 MAPPED POINT PATTERN DATA

Figure 1.3 shows a mapped pattern consisting of 3, 605 Beilschmiedia
trees in a rectangular stand of tropical rain forest at Barro Colorado
 Mapped point pattern data  5

Figure 1.3Top: positions of Beilschmiedia trees in a 1,000 by 500 metre

stand in Barro Colorado Island, Panama. Bottom: norm of elevation (in
metres) gradient in the stand.

Island, Panama. These data were prepared for R by R. Waagepetersen

and taken from a larger data set described in [4].
We will only be interested in maps where the mechanism that gener-
ated the points is of interest. For instance, since the map of centroids of
the North Carolina counties discussed in the previous section is purely
artificial and has no bearing on the abundance of SIDS cases, there is no
point in studying it. For genuine mapped point pattern data, research
questions tend to focus on the arrangement of the points, in particular,
on trends and interactions.
Returning to Figure 1.3, it is clear at first glance that the trees are
not distributed over the plot in a uniform way. Rather, they seem to
be concentrated in specific regions. Possible explanations could include
local differences in soil quality or the availability of nutrients, differences
in the terrain, or traces of a planting scheme. To quantify and test non-
homogeneity, the forester may use quadrats, that is, a partition of the
stand in disjoint spatial bins, and apply classical statistical dispersion
tests to the quadrat counts. It might also be of interest to test whether
the counts follow a Poisson distribution.
6  Introduction

Since Barro Colorado island has been studied extensively over the
past century, a lot is known about the terrain. The image in the lower-
most panel of Figure 1.3 displays the norm of the elevation gradient.
Visually, it seems that a steep gradient aligns well with a high tree
intensity, a correlation that the forester may be interested in quantifying
by means of a spatial generalised linear regression model.
The steepness of the terrain is only one factor in explaining the
mapped pattern. A cluster in the left part of the stand, for example,
is rich in trees, even though the terrain there is not steep at all. Addi-
tionally, there could be interaction between the trees due to competition
for nutrients or sunlight or because of seed dispersion patterns that the
forester may try to capture in a model.
Finally note that additional measurements might be taken at each
tree location, for example the number of stems, the size of the crown or
the diameter at breast height, but we will not pursue this topic further.

1.4 PLAN OF THE BOOK

Chapter 2 is devoted to gridded data such as the coal ash measurements
displayed in Figure 1.1. The mathematical definition of a random field is
given before specialising to Gaussian random fields. Such random fields
are convenient to work with since their distribution is fully described by
the mean and covariance functions. Next, various types of stationarity
are discussed and shown to be equivalent for Gaussian random fields.
The celebrated Bochner theorem provides a spectral representation for
continuous covariance functions.
The second part of the chapter is dedicated to spatial interpolation.
First, the semi-variogram and its empirical counterpart are introduced
to quantify the local interaction structure in the data. A simple kriging
procedure is developed that is appropriate when both the mean and the
semi-variogram are known explicitly. It is shown that this procedure re-
duces to a Bayes estimator when the random field is Gaussian. In the
last sections of the chapter, the strong assumptions on mean and semi-
variogram are relaxed. More precisely, ordinary kriging is the name given
to spatial interpolation when the mean is constant but unknown. Uni-
versal kriging is apt when explanatory variables are available to define
a spatial regression for the mean.
Chapter 3 is concerned with areal unit data such as the infant death
counts shown in Figure 1.2. Special attention is given to autoregression
models, including Gaussian and logistic ones. It is shown that, provided
 Plan of the book  7

a positivity condition holds, the distribution of such models is fully de-

scribed by their local characteristics, that is, by the family of the condi-
tional distributions of the measurement at each areal unit given those at
other units. When these local characteristics are truly local in the sense
that they depend only on the neighbours of the areal unit of interest,
the random field is said to be Markov. Using the theory of Gibbs states,
it is proved that the joint probability density of a Markov random field
can be factorised in interaction functions on sets of mutual neighbours.
The second part of the chapter is devoted to statistical inference,
in particular estimation of the model parameters. First, the maximum
likelihood equations are derived for a Gaussian autoregression model.
For most other models, the likelihood is available only up to a parame-
ter dependent normalisation constant. Several techniques are discussed,
including maximum pseudo-likelihood and Monte Carlo maximum like-
lihood estimation. The chapter closes with two examples of hierarchical
modelling and inference, image segmentation and disease mapping.
The last chapter, Chapter 4, features mapped point pattern data
such as the map of trees in Figure 1.3. The formal definition of a point
process is given before specialising to Poisson processes. These processes
are convenient to work with because of the lack of interaction between
their points, and the fact that their distribution is fully described by the
intensity function. Next, the moment measures and associated product
densities are defined for general point processes, together with their em-
pirical counterparts. Various concepts of stationarity are also discussed.
The remainder of the chapter is restricted to finite point processes.
Following similar lines as those laid out in Chapter 3, a family of condi-
tional distributions is defined on which a Markov property can be based
and a factorisation of the joint probability density in terms of interac-
tion functions defined on sets of mutual neighbours is seen to hold. A
maximum likelihood theory is developed for Poisson processes, whilst
the maximum pseudo-likelihood and Monte Carlo maximum likelihood
methods apply more generally. Minimum contrast techniques can be used
for point processes, including Cox and cluster processes, for which the
likelihood is intractable. An application to cluster centre detection con-
cludes the chapter.
Each chapter also contains worked examples and exercises to illus-
trate, complement and bring the theory into practice. In order to make
the book suitable for self-study, solutions to selected exercises are col-
lected in an appendix. The chapters close with pointers to the original
 8  Introduction

sources of the results, in so far as it was possible to trace them, and to

more specialised and elaborate textbooks for further study.
The calculations in this book were done using the R-language, a free,
open source implementation of the S programming language created by
J.M. Chambers [5]. R was created in the 1990s by R. Ihaka and R.
Gentleman and is being developed by the R Development Core Team
currently consisting of some twenty people. For an introduction, we refer
to [6]. An attractive feature of the R-project is that it comes with a great
many state of the art packages contributed by prominent researchers.
The current list of packages is available at the site cran.r-project.org.
A bit of a warning, though. Packages come with absolutely no warranty!
Of course, it is also possible to write one’s own functions and to load
C-code.

REFERENCES
[1] M. Gomez and K. Hazen (1970). Evaluating sulfur and ash distribution in coal
seams by statistical response surface regression analysis. U.S. Bureau of
Mines Report RI 7377.

[2] D. Atkinson (1978). Epidemiology of sudden infant death in North Car-

olina: Do cases tend to cluster? North Carolina Department of Human
Resources, Division of Health Services Public Health Statistics Branch
Study 16.

[3] M.J. Symons, R.C. Grimson and Y.C. Yuan (1983). Clustering of rare events.
Biometrics 39(1):193–205.

[4] S.P. Hubbell and R.B. Foster (1983). Diversity of canopy trees in neotropical
forest and implications for conservation. In: Tropical Rain Forest: Ecology
and Management. Edited by S. Sutton, T. Whitmore and A. Chadwick.
Oxford: Blackwell.

[5] R.A. Becker, J.M. Chambers and A.R. Wilks (1988). The New S Language.
Pacific Grove, California: Wadsworth & Brooks/Cole.

[6] P. Dalgaard (2008). Introductory Statistics with R (2nd edition). New York:
Springer-Verlag.
 CHAPTER 2

Random field modelling

and interpolation

2.1 RANDOM FIELDS

Climate or environmental data are often presented in the form of a map,
for example the maximum temperatures on a given day in a country,
the concentrations of some pollutant in a city or the mineral content in
soil. In mathematical terms, such maps can be described as realisations
from a random field, that is, an ensemble of random quantities indexed
by points in a region of interest.
Definition 2.1 A random field is a family X = (Xt )t∈T of random
variables Xt that are defined on the same probability space and indexed
by t in a subset T of Rd .
Let us consider a finite set t1 , . . . , tn ∈ T of index values. Then the
random vector (Xt1 , . . . , Xtn ) has a well-defined probability distribu-
tion that is completely determined by its joint cumulative distribution
function

Ft1 ,...,tn (x1 , . . . , xn ) = P(Xt1 ≤ x1 ; · · · ; Xtn ≤ xn ),

where xi ∈ R for i = 1, . . . , n. The ensemble of all such joint cumulative

distribution functions with n ranging through the natural numbers and
t1 , . . . , tn through T constitute the finite dimensional distributions or
fidi’s of X. Together, they uniquely define the probability distribution
of X.
The proof relies on Kolmogorov’s consistency theorem which states
the following. Suppose that for every finite collection t1 , . . . , tn , we have

9
10  Random field modelling and interpolation

a probability measure μt1 ,...,tn on Rn with joint cumulative distribution

function Ft1 ,...,tn . If this family of fidi’s is symmetric in the sense that

Ftπ(1) ,...,tπ(n) (xπ(1) , . . . , xπ(n) ) = Ft1 ,...,tn (x1 , . . . , xn )

for all n ∈ N, all x1 , . . . , xn ∈ R, all t1 , . . . , tn ∈ T and all permutations

π of (1, . . . , n), and consistent in the sense that

lim Ft1 ,...,tn (x1 , . . . , xn ) = Ft1 ,...,tn−1 (x1 , . . . , xn−1 ),

xn →∞

for all n ∈ N, all x1 , . . . , xn−1 ∈ R and all t1 , . . . , tn ∈ T , then there

exists a random field X whose fidi’s coincide with those in F .
In summary, in order to define a random field model, one must specify
the joint distribution of (Xt1 , . . . , Xtn ) for all choices of n and t1 , . . . , tn
in a consistent way. In the next section, we will assume that these joint
distributions are normal, and show that in that case it suffices to specify
a mean and covariance function. For this reason, Gaussian models are
widely used in practice. Alternative modelling strategies may be based
on transformations, linear models, series expansions or deterministic or
stochastic partitions of T , of which we present a few simple examples
below.

Example 2.1 Fix n ∈ N and consider a partition A1 , . . . , An of T . More

precisely, the Ai are non-empty, disjoint sets whose union is equal to T .
Let (Z1 , . . . , Zn ) be a random n-vector and write

n
Xt = Zi 1{t ∈ Ai }
i=1

for all t ∈ T . In other words, the random surface defined by X is flat

on each partition element Ai . The value set of the Zi may be finite,
countable or a subset of R. In all cases, Xt is a linear combination of
random variables and therefore a random variable itself.

Example 2.2 Fix n ∈ N and let fi : T → R, i = 1, . . . , n, be a set of

functions. Let (Z1 , . . . , Zn ) be a real-valued random n-vector and write

n
Xt = Zi fi (t), t ∈ T.
i=1

Then Xt is a well-defined random variable. The fi may, for example, be

harmonic or polynomial base functions, or express some spatial charac-
teristic of interest.
 Gaussian random fields  11

One may also apply transformations to a random field to obtain new

ones.

Example 2.3 Let X = (Xt )t∈T be a random field and φ : R → R a

measurable function. Then φ(X) = (φ(Xt ))t∈T is also a random field.
Note that the supports of the random variables Xt and φ(Xt ) may differ.
The transformation φ : x → exp(x), for instance, ensures that φ(Xt )
takes positive values.

2.2 GAUSSIAN RANDOM FIELDS

Recall that a random variable is normally or Gaussian distributed if it
has probability density function
 
1 (x − μ)2
f (x) = exp − , x ∈ R,
σ(2π)1/2 2σ 2

with σ 2 > 0 or if it takes the value μ with probability one, in which case
σ 2 = 0. The constant μ ∈ R is the mean, σ 2 the variance.
Similarly, a random vector X = (X1 , . . . , Xn ) has a multivariate
normal distribution with mean vector m = (EX1 , . . . , EXn ) ∈ Rn and
n × n covariance matrix Σ with entries Σij = Cov(Xi , Xj ) if any linear

combination a X = ni=1 ai Xi , a ∈ Rn , is normally distributed.
The normal distribution plays a central role in classical statistics. In
a spatial context, we need the following analogue.

Definition 2.2 The family X = (Xt )t∈T indexed by T ⊆ Rd is a Gaus-

sian random field if for any finite set t1 , . . . , tn of indices the random
vector (Xt1 , . . . , Xtn ) has a multivariate normal distribution.

By the definition of multivariate normality, an equivalent charac-


terisation is that any finite linear combination ni=1 ai Xti is normally
distributed.
The finite dimensional distributions involve two parameters, the
mean vector and the covariance matrix. The entries of the latter are
Cov(Xti , Xtj ), i, j = 1, . . . , n. Define the functions

m : T → R; m(t) = EXt

and
ρ : T × T → R; ρ(s, t) = Cov(Xs , Xt ).
12  Random field modelling and interpolation

They are called the mean and covariance function of X. If we know m

and ρ, we know the distributions of all (Xt1 , . . . , Xtn ) , t1 , . . . , tn ∈ T .
However, not every function from T × T to R is a proper covariance
function.

Example 2.4 Examples of proper covariance functions include the fol-

lowing.

1. The choices T = R+ = [0, ∞), m ≡ 0 and

ρ(s, t) = min(s, t)

define a Brownian motion.

2. For m ≡ 0, β > 0, and

1
ρ(s, t) = exp (−β||t − s||) , s, t ∈ Rd ,
2β
we obtain an Ornstein–Uhlenbeck process. The function ρ is alter-
natively known as an exponential covariance function.

3. For β, σ 2 > 0, the function

 
ρ(s, t) = σ 2 exp −β||t − s||2 , s, t ∈ Rd ,

is the Gaussian covariance function.

4. Periodicities are taken into account by the covariance function

ρ(s, t) = σ 2 sinc(β||t − s||), s, t ∈ Rd ,

for β, σ 2 > 0 defined in terms of the sine cardinal function

sinc(x) = sin(x)/x for x = 0 and 1 otherwise. Note that the corre-
lations are alternately positive and negative, and that their absolute
value decreases in the spatial lag ||t − s||.

Proposition 2.1 The function ρ : T × T → R, T ⊆ Rd , is the covari-

ance function of a Gaussian random field if and only if ρ is non-negative
definite, that is, for any t1 , . . . , tn , n ∈ N, the matrix (ρ(ti , tj ))ni,j=1 is
non-negative definite.
 Stationarity concepts  13

In other words, for any finite set t1 , . . . , tn , the matrix (ρ(ti , tj ))ni,j=1
should be symmetric and satisfy the following property: for any a ∈ Rn ,

n 
n
ai ρ(ti , tj )aj ≥ 0.
i=1 j=1

Proof: “⇒” Since (ρ(ti , tj ))ni,j=1 is the covariance matrix of (Xt1 , . . . ,

Xtn ) , it is non-negative definite.
“⇐” Apply Kolmogorov’s consistency theorem. To do so, we need
to check the consistency of the fidi’s. Define μt1 ,...,tn to be a multivari-
ate normal with covariance matrix Σ(t1 , . . . , tn ) having entries ρ(ti , tj ).
By assumption, Σ(t1 , . . . , tn ) is non-negative definite so that μt1 ,...,tn is
well-defined. The μt1 ,...,tn are also consistent since they are symmetric
and the marginals of normals are normal with the marginal covariance
matrix. 

Example 2.5 For T = Rd , set ρ(s, t) = 1{s = t}. Then ρ is a

proper covariance function as, for any t1 , . . . , tn ∈ T , n ∈ N, and all
a1 , . . . , an ∈ R,

n 
n 
n
ai ρ(ti , tj )aj = a2i ≥ 0.
i=1 j=1 i=1

More can be said for Gaussian random fields under appropriate sta-
tionarity conditions, as will be discussed in the next two sections.

2.3 STATIONARITY CONCEPTS

Throughout this section, the index set will be T = Rd .

Definition 2.3 A random field X = (Xt )t∈Rd is strictly stationary if

for all finite sets t1 , . . . , tn ∈ Rd , n ∈ N, all k1 , . . . , kn ∈ R and all
s ∈ Rd ,

P(Xt1 +s ≤ k1 ; · · · ; Xtn +s ≤ kn ) = P(Xt1 ≤ k1 ; · · · ; Xtn ≤ kn ).

Let X be strictly stationary with finite second moments EXt2 < ∞

for all t ∈ Rd . Then

P(Xt ≤ k) = P(Xt+s ≤ k)
14  Random field modelling and interpolation

for all k so that Xt and Xt+s are identical in distribution. In particular,

EXt = EXt+s . We conclude that the mean function must be constant.
Similarly,

P(Xt1 ≤ k1 ; Xt2 ≤ k2 ) = P(Xt1 +s ≤ k1 ; Xt2 +s ≤ k2 )

for all k1 , k2 so that the distributions of (Xt1 , Xt2 ) and (Xt1 +s , Xt2 +s )
are equal. Hence Cov(Xt1 , Xt2 ) = Cov(Xt1 +s , Xt2 +s ). In particular, for
s = −t1 , we get that

ρ(t1 , t2 ) = ρ(t1 + s, t2 + s) = ρ(0, t2 − t1 )

is a function of t2 − t1 . These properties are captured by the following

definition.

Definition 2.4 A random field X = (Xt )t∈Rd is weakly stationary if

• EXt2 < ∞ for all t ∈ Rd ;

• EXt ≡ m is constant;

• Cov(Xt1 , Xt2 ) = ρ(t2 − t1 ) for some ρ : Rd → R.

Since a Gaussian random field is defined by its mean and covari-

ance functions, in this case the reverse implication (weak stationarity
implies strict stationarity) also holds. To see this, consider the ran-
dom vector (Xt1 , . . . , Xtn ) . Its law is a multivariate normal with mean
vector (m, . . . , m) and covariance matrix Σ(t1 , . . . , tn ) with ij-th entry
ρ(tj − ti ). The shifted random vector (Xt1 +s , . . . , Xtn +s ) also follows a
multivariate normal distribution with mean vector (m, . . . , m) and co-
variance matrix Σ(t1 + s, . . . , tn + s) whose ij-th entry is ρ(tj + s − (ti +
s)) = ρ(tj − ti ) regardless of s. We conclude that X is strictly stationary.
Defintion 2.4 can easily be extended to random fields that are defined
on a subset T of Rd .

Example 2.6 Let (Xt )t∈Rd be defined by


d
Xt = (Aj cos tj + Bj sin(tj )) , t = (t1 , . . . , td ) ∈ Rd ,
j=1

where Aj and Bj , j = 1, . . . , d, are independent random variables that

are uniformly distributed on [−1, 1]. Then Xt is not strictly stationary.
Another random document with
no related content on Scribd:
we had better only have the kitchen fire in the mornings; but I don't like to
think of you going out to work cold."

"I don't feel the cold," said Luke. "We certainly must make a difference
somehow. Discuss it with mother and see if she can't help. She knows we are
rather in low water."

"I don't see how we can give away so much as you do Luke," said Rachel.
"We really could not afford that £5 that you gave for the heating of the Church
for instance."

Luke looked worried.

"I have always somehow managed to do my part in that way," he said. "I can't
bear not setting an example in giving."

"No, it's horrid," said Rachel. And yet she felt strongly, that if by giving away
money he was deprived of his much needed yearly holiday the work itself
would suffer.

After he had gone, the subject that had caused their talk and had brought to
light their poverty forced itself again upon her.

No holiday! All the summer in this tiny stuffy little house away from the flowers
and the breezy wind. How could she bear it herself, and still more how could
Luke go on working all day and the greater part of the evening in the terrible
heat, which was making her feel giddy and faint already.

She hurried into the drawing-room which was without sun and threw the
window open. Then she looked for her weekly bill books and sat down to
examine them. She saw they were higher than she supposed they ought to
be, but she did not see how she could economise with a man in the house. If it
were only herself and Polly they could do on less; but Luke, though he might
not notice that he was having less would soon reap the consequence and feel
limp.

Rachel leant her arms on the bureau and her chin on the palms of her hands
and gazed out Of the window. How could they manage to get away for a week
if no longer?

Unfortunately Heatherland was impossible.

Her mother found herself so impoverished that she was selling her house and
was on the point of going into a much smaller one in the village. Rachel had
felt very sad when Sybil had written to tell her the news. That her mother, at
her age, should have to uproot again would be a real trial; evidently her family
were in financial difficulties too.

Rachel began to think over her belongings and wondered if she could not sell
some of her wedding presents. There was the pearl necklace that an uncle
who was dead had given her. It was of little use to her now, and in her present
mood she felt that a breath of sea air would compensate for the loss of any
number of pearls.

Yes, she would certainly sell her pearls. She wondered if the day would come
when she would be reduced to selling many of her possessions. It looked like
it. It was a terrible shame that livings should be so small that the very
necessities of life should have to be done without. Well anyhow she would sell
her pearls and not tell Luke till it was done. She would get a cousin of hers to
do the transaction for her. She knew she might be cheated and it would be no
good for Luke to try and sell them. He was no business man and would
without doubt be contented with half their value. No, she would write to her
cousin. They simply must go away somewhere this summer.

She wrote the letter to her cousin and got Polly to run round to the post with it.
Then she began to wonder if she ought not to do what Luke had suggested;
ask the advice of Mrs. Greville. But she was saved the trouble, for late in the
morning her mother-in-law came round to see her.

"Luke tells me you are rather worried about the expenses," she said, "and I
am wondering if I can help you. Shall I look through your books and see what
you could do without?" and as she saw Rachel flush she added, "It is not at all
surprising my dear. Of course you have never been used to economise. I
hope you don't me an interfering old woman," she added kindly, as she saw
signs of distress on Rachel's face.

"It's very good of you," said Rachel; but she bit her lip feeling humiliated in the
extreme.

Mrs. Greville was not long in discovering things which would have to be done
without. For instance, she explained to Rachel she could make quite nice
puddings without eggs. Considering the expense of eggs, 4d each, it was
ruinous to follow the cookery books which prescribed more than one in quite
simple puddings. A great deal of money had been thrown away on
unnecessary eggs and they mounted up at once. Then it was much better in
these days to have margarine rather than butter. She never used anything but
margarine herself, and really you would not know the difference.

Rachel sat by her side smiling. Not being a housekeeper all these economies
seemed so paltry to her, and yet she knew they were necessary. She had of
her own will married a poor clergyman, and must bear the consequences. And
Mrs. Greville was being very kind; and giving as little pain as possible. Her
feelings towards Rachel had somewhat changed since that Sunday on which
she had gathered the girls' choir together. It had struck her mother-in-law as a
sporting action on her part and had pleased her. And now, for the sake of her
son, she was very anxious not to hurt his wife's sensitive spirit more than
necessary, but she had no idea how galling the whole thing was to her pride.

And Rachel was bent on her not guessing it. So she sat by her side smiling,
and watching her mother-in-law making notes for her as to the things that
were really necessary to have and those which were mere luxuries.

"I fear you must give up all luxuries, I hope Luke told you how poor he is,
when he asked you to share his poverty with him?" she said laughing. Rachel
joined in the laugh.

"We had much more interesting things to talk about," she answered. "Luke's
mind does not run on such matters as eggs and margarine."

"That's true," said Mrs. Greville. "The fact is, that some of the poorest people
are the clergy. It ought not to be so. They should not have to worry about eggs
and margarine as you say, they have so much more important things to think
of and they should be spared that. Besides they are expected to help in every
bit of work that goes on in the parish. Unless they have property of their own
the worry of pounds, shillings, and pence, weighs them down. Happily, as you
say, Luke does not worry himself about those kind of things, but then he has a
wife and mother to worry for him. If he had not he would have less time and
strength to think of his people. I don't suppose many realise how the clergy
suffer from poverty, for they suffer in silence."

"Well I hope Luke won't ever be reduced to wearing a coat green from age, as
a poor man in our part of the world at home has to do. Not that I suppose
Luke would notice if his coat were all colours of the rainbow."

Mrs. Greville laughed, and said goodbye while at the door she turned back to
say:
"Remember, not so many eggs, and margarine instead of butter. You'll find
that makes a difference very soon."

CHAPTER XV.
RACHEL'S PEARLS.

Rachel took out her pearls from the jewel box and looked at them. They were
certainly very beautiful. She had not worn them since her wedding day; and
she did not see any chance of wearing them again.

In case the parting with them should grow a little hard she packed them up
quickly and went to the post to register them.

It seemed to her as if she was parting with another link of the old life. But after
all what did that matter! She had Luke; and it was true what Gwen had said,
that Luke compensated for the loss of all else. Besides which, she knew that
they both needed a change and rest, and certainly sea breezes were of more
value just then than pearls locked up in her jewel case could be. Nevertheless
it cost her something to part with their beauty. It was not so much their value
that she had thought of as their beauty; and more than once she had taken
them out simply to have the pleasure of seeing something very lovely. There
was so little beauty surrounding her that she revelled in the sight of her pearls.
 RACHEL TOOK OUT HER PEARLS—AND LOOKED AT THEM.

It was some time before the cousin, to whom she had sent them, wrote and
told her that he believed they would fetch a very good price; and one morning
at breakfast she opened a letter that was lying on the table beside her, and a
cheque, much larger than she had hoped for, fell out.

Her exclamation of surprise and pleasure caused Luke to look up from the
paper he was reading.

"What is it?" he asked.

"Something delightful," she answered. "We can now go for our holiday."

"What do you mean? Have you come in for a fortune?"

"I feel as if I had in our present strait," said Rachel still looking down at the
cheque in her hand. "Would you like to see what I have?" She held it out to
him.

Money meant little to Luke, except that it enabled him to carry out plans on
which he had set his heart. And there was a plan of his that sprang at once
into his mind as he looked at the cheque. His face glowed.

"Where did it come from?" he asked, astonished and delighted.

"I sold my pearls."

The smile on his face faded for a moment.

"I did not know you had any, I have never seen them."

"I have never shown them to you as I know that that kind of thing is not in your
line, you don't care for jewels; but I wore them on my wedding day. I hope you
saw them then."

"No, I only saw you. But why didn't you tell me what you were doing?"

"Because I was afraid you would not let me part with them. Uncle Joe gave
them to me, and I was fond of him."

"I shouldn't have prevented you from parting with them. What is the good of
pearls?"

"The good! Why their beauty is their good. They are gifts from God just as
everything else is that is lovely, and of good report. Don't despise them.
Besides," she said, feeling a little sore, "Uncle Joe gave them to me, and I
loved him." Then she added, determined not to give way to any feeling of
disappointment, "and now we can think about our holiday. Where shall we
go?"

She was folding the cheque up and putting it again in the envelope of her
letter. But on noticing that Luke did not answer her question, she glanced up
and found him looking out of the window with a dreamy happy smile on his
face. He was evidently thinking of the holiday. Perhaps his thoughts had flown
to Southwold and the moon's silver pathway on the sea. The happiness
displayed in his expression of face made her feel that the small self-denial that
she had exercised was well worth while.
"I do believe," he said, still looking out of the window, "that at last my dream
will be fulfilled."

"What is your dream?" asked Rachel. She had been right. He was evidently
dreaming of Southwold.

"Why, to put electric light in St. Marks. Think how attractive and bright it would
make the place. I never thought I should be able to do it. How much do you
think it would cost?"

Rachel was silent from astonishment and disappointment.

Then she said slowly:

"You don't suppose, do you Luke, that I have sold my pearls to be able to do
for the Church what the people are far better able to afford to do than we are?
You seem to forget that we are really poor, much poorer than many of the
congregation. No, the first thing we must do," she said decidedly, "is to pay
our bills and to start afresh, and then to go for a holiday."

Rachel's tone of voice was so decided that he turned and looked at her in
surprise.

"But," he said, "it is God's work that you are refusing."

"No," she answered, "I don't think it is. To provide the luxury of electric light in
a place where the gas is quite good and sufficient, seems to me to be not so
much doing the Will of God as paying our debts, and going for a holiday,
which will give you strength to do His work better."

Luke was silent.

"Besides," she added in a lower voice and smiling, "you forget next January."

Luke looked mystified. Then noticing the expression of his wife's eyes, he
remembered.

"I forgot for the moment," he said. "You are quite right."

At the thought of January Rachel's face had lost all the surprise and
disappointment as she looked down again at the envelope containing the
cheque. Then she suddenly drew it out with a gay laugh and flourished it in his
face.
"Happily it's mine," she said, "and you can't get at it."

CHAPTER XVI.
THE CHURCH COUNCIL.

"I know you won't give way," said Rachel, as she helped her husband into his
coat.

Luke did not answer at once.

"If you remain firm this time you will find it easier the next."

"I don't intend to give way," said Luke gravely. "It would be fatal."

"Yes, and you would never forgive yourself." Luke buttoned up his coat and
looked at his wife.

"We should hear the same excuse that that old woman gave for not attending
her Parish Church," he said smiling.

"What was that?"

"That she had been too ill for the last half year to come to the services but she
was proud to say she had not missed a single Whist Drive."

"Oh you mustn't give way," repeated Rachel earnestly. Then she added, laying
a hand on his arm, "You know what I shall be doing."

When Luke smiled his face was transfigured.

He smiled now.

"You will be doing the part of Moses," he said. "I am not sure that I should not
fail if it were not for that." Then he opened the door and was gone.

Rachel stood where he had left her looking absently at the door which he had
shut after him.

That the Church Councils were not always happy meetings she knew. Luke
said the right people were not on it. There were some really in earnest, but
these often seemed afraid of speaking. Those who spoke the most often and
the loudest were those who wanted to make their Church the most popular in
the town by way of parish dances, whist drives, etc. Luke, Rachel knew, had
always stood out against such methods of work, but people were growing
persistent, and the subject was to come up again this evening. His last words
had surprised her and made her anxious, for they showed her that his
resistance was growing weaker and that he felt himself in real danger of giving
way.

She knew what giving way would mean to him. It would lie on his heart like
lead. He would not look for blessing so expectantly and hopefully in his parish
if once his church began to cater for the amusement of his people instead of
putting its full strength into the spiritual work. By this time Rachel knew her
husband so well, that she felt sure that he would grow melancholy and
depressed, and his work would be robbed of zeal and happiness in
consequence.

It was not as if he had any doubts as to the wrongfulness of such methods for
Church work. He absolutely disapproved of them and had made his opinions
known. If he gave way or countenanced such proceedings, in the least,
people would cease to believe in him.

Well there was one thing she could do to help him to be strong; she would go
and do it. And while Luke was wrestling with his Church Council, Rachel was
wrestling in prayer.

Then she went down stairs to listen for his footstep.

When she heard it, it did not inspire her with hope. Luke came in quietly and
made his way slowly into the drawing-room where he knew he would find her.

"Well?" she said.

He took a seat on the sofa near her.

"I have gained my point."

Rachel's eyes shone.

"What good news," she said greatly relieved. "Why do you look so
melancholy?"

"Because though I have gained my point I have lost four of the most regular
members of the congregation. They walked out of the room."

"Oh well, that is not half so bad as if they had gained the victory," said Rachel
cheerfully. Then Luke looked lovingly down at his wife.

"I doubt if I should have taken such a strong attitude if it had not been for you.
I knew you were praying."

"Yes, I was praying."

"The knowledge of that helped me enormously. The four who resigned have
been my most loyal supporters and I can tell you it was hard to stand out
against them. They have been so exceedingly kind to me ever since I have
had the church. It was this fact that made it so difficult. Besides I love peace."

"Peace with honour, but not without."

"That's just it. I felt that my Master's honour was at stake."

"I can't tell you how thankful I am," said Rachel. And under his wife's influence
Luke regained something of his usual spirits. But Rachel had only heard
Luke's side of the question. The next day she was inundated with callers.

"I suppose you know," said Mrs. Moscombe, the wife of the owner of the
principal shop in the parish, "that the Vicar got his way by one vote only. I own
when I heard all the arguments in favour of opening the Hall for such
purposes I began to wonder if we ought not to do what the neighbouring
churches have done to attract the young people."

"I don't think that kind of thing attracts people to Church," said Rachel.

"But as my husband says," continued Mrs. Moscombe, "it saves the boys and
girls from going to worse places. Surely that is the work of the Church."
"It may save them for two or three days, possibly; but it really leads them to go
to unwise places of amusement in the long run, and I know my husband feels
very strongly that the Church loses its spiritual power if it goes in for catering
for amusements."

"But then, dear Mrs. Greville, your husband, forgive me for saying so, is rather
peculiar in his views. He scarcely moves with the times and isn't up to date as
they say."

Rachel flushed.

"The times are not so particularly good that one should wish to move with
them," she said. "I am very thankful that my husband does what he considers
right without swerving or moving with the times."

Rachel was glad to see the last of her visitor, but had hardly said goodbye
when the door opened to admit Mrs. Stone.

"Of course," she said as she took a seat, "your husband has told you all about
last night. I admired him immensely. He didn't give way an inch though the
majority were really against him."

"But anyhow he had a majority of one."

Mrs. Stone laughed.

"And he would not have had that if it had not been for me. I didn't agree with
him in the least! I must tell you, but I voted for his views as I always feel he is
such a good man that he probably knows what is for the good of his people
better than I do. If it had not been for my vote, he would himself have had to
give the casting vote."

"I am very disappointed that you don't agree with us," said Rachel.

"Well I do believe in people keeping up with the times, and girls and boys are
crazy now for dancing and cards. You can't get them if you don't give way.
Things have changed so much since our fathers' times."

Rachel was silent. She felt depressed. She quite expected people like Mrs.
Moscombe and others who had called to see her, and who did not profess to
be religious, to misunderstand Luke's action, but it was a blow to find that her
friend Mrs. Stone also disagreed with him.

But the last caller was the most trying of all.

Rachel heard her mother-in-law open the front door and walk heavily across
the little hall.

"What has Luke been doing?" she exclaimed almost before she was in the
room. "I hear he has quite estranged the four best supporters of the Church."
She looked at Rachel as if she were to blame.

"It was about the amusement question," said Rachel. "He put down his foot at
the proposal to introduce them into the work of the parish."

"Well I call it remarkably silly of him. It is a matter of very little importance and
certainly not worth wrangling over. I am quite thankful I am not a member of
the Council. I could not have voted against my son, but I should have felt very
vexed at being a party to such a loss to the Church."

"You mean?"

"I mean losing his four best financial supporters, and those who give the most
to the Easter offerings. Who is he to look to now, I should like to know? And if
he only waited to consider the state of his own finances and the expense of
food, (eggs are still fourpence a piece), he would not have made such a fatal
mistake."

Rachel was silent, but she disagreed with every word her mother-in-law had
spoken. Then after a pause during which Mrs. Greville tied and untied her
bonnet strings in her agitation, she said:

"I don't suppose any consideration respecting finance would weigh with Luke
against doing what he thinks right."

"My dear, young men often make a fatal mistake in going their own way,
thinking that youth must know better than age. Think of those four gray haired
men who know more of the world than Luke, being set at nought like that. I
have never known Luke to make such a mistake. If he had only consulted me
before he had acted."

Then Rachel spoke.

"But don't you see how noble it was of him to keep to what he felt right even
though he must have known what the result would be. I own am proud of him,
and should have been bitterly disappointed if he had given way. I am sure he
did the right thing."
Mrs. Greville looked at her son's wife and could not but admire the way she
stood up for her son's folly, (as she considered it). There was an expression
on her face that any mother-in-law would have been pleased to see on the
face of her son's wife. But for all that she felt it incumbent on her to give her a
snub.

"I daresay," she said, "that you admire him. So would most young girls who
only look for actions without weighing their cost. We all admire a man who is
not afraid to speak out. But when it comes to flouting those who have been
kind and considerate, and who never hesitate to give money for the work, it is
a different matter. Luke has done a bad thing for the parish by his action of
last night."

"People would never have believed in his convictions again if he had given
way," said Rachel.

"Well now, don't you go and encourage him in that kind of thing," said Mrs.
Greville. "I hope that you recognise the fact that Luke is not a paragon of
wisdom, neither can any one turn him from what he imagines his duty. But he
must remember that he now has a wife to support. He not only will stubbornly
stick to his point even when it means losing money for the work of the parish,
but will give away every penny he possesses without a thought of the
consequences. I daresay you have found that out."

Rachel laughed, thinking of the pearls.

"Well, am I not right?"

"Luke is the most generously minded man that I have ever come across," said
Rachel.

And then Mrs. Greville gave her a kiss. She could not resist it; though she
knew that her action would startle her daughter-in-law.

"My dear," she said, "I do believe that you love that boy of mine as much as I
do."

Rachel was tempted to answer "a great deal more," but forbore, only returning
the kiss with warmth. She was getting almost fond of Mrs. Greville.

"Well, you see," she said with a smile, "he is my husband."

"But that does not always follow I am sorry to say. Wives are generally very
quick in seeing and resenting faults in their husbands. And much as I love my
dear boy I see a great many in him."

"But there are more virtues after all," was Rachel's answer, "and to return to
the subject of whist drives, it is perfectly true what Luke quoted to me from
some speech yesterday. 'The Church has so little power with the world
because the world has so much power over the Church.' Don't you agree?"

"Well perhaps it is so. But when a man's bread and butter is concerned and
when the Church funds are low, I own I feel it is not the time to be too
particular."

"I am afraid I don't agree with you a bit."

"I don't suppose you do; that is because Luke has imbued you with his ideas
of right and wrong."

"Luke has a very high ideal," said Rachel, "and I am trying to live up to it."

And Mrs. Greville went away thinking to herself, "I only hope that Luke
realises what a devoted wife he has. I don't believe he does."

CHAPTER XVII.
LUKE IS DISCOURAGED.
The four members who had left the Church Council when the vote went
against Whist Drives for Church purposes, did not leave the Church. They
valued their Vicar too much to do so suddenly; but they were thoroughly vexed
at the decision arrived at.

It had been a blow to Luke to find that he only had a majority of one. He had
hoped that his congregation had felt with him in the matter, and finding how
strong the stream was towards such means of increasing the popularity of the
Church, depressed him not a little.

Moreover, he felt bound, greatly against his will, to preach against such
methods and to give his reasons for so doing; and though some respected
him for his courage, there were others who resented it. To Rachel, the Sunday
on which he mentioned the matter was a most painful day; though she was
glad that Luke had spoken out on what was much on his mind.

"I am beginning to think," he said, on sitting down to dinner after the service,
"that my time in the parish is about over. It seems to me my influence for good
is not strong enough. It wants a stronger man than I am here."

"You are tired," said Rachel. "That's what is making you downhearted. Last
week was such a very heavy one for you. In a day or two you will see things
differently."

Luke smiled unbelievingly.

"What I should like, and in fact what I have always longed for," he said, "is a
Church in London. Though I doubt if such an honour will ever come my way. I
am not a big enough man to be trusted with a London parish."

"London!" cried Rachel. "Oh Luke, I should hate it of all things. Besides you
must be a man strong in body as well as in soul to work a London parish
satisfactorily. I should be very sorry to see you undertake such a work."

"I'm quite strong enough," said Luke. "The only things that try me are the petty
quarrels and vexations of such a parish as this. I heard this morning that Went
and Ethers have fallen out, and on a ridiculously small matter. I fancy
everything would be larger and more important in London. It is just the petty
matters that worry me."

"Human nature is the same everywhere. I expect you would find small souls in
a London parish just as you do here."
"Would you very much object to London?" asked Luke. "Not that there is the
slightest chance of me being offered a Church there. But it is the dream of my
life. Fancy working in the very hub of the Universe. I should revel in it."

"The work would be enormous, unless you had several curates. And you know
how difficult they are to find now-a-days."

"I shouldn't mind the work. The more the better, so long as it is not spoilt by
bickerings and quarrellings. Should you very much dislike it?"

"Intensely. I don't feel in my present mood, as if I could endure it." Then seeing
a look of disappointment on her husband's face, she added, "But where thou
goest I will go, you know that."

"Yes, I have no doubt of that," he answered.

And his longing for London increased during the next few months. It was a
time of great disappointment for him. When he had first come to Trowsby, he
had had the warmest of welcomes, and the largest congregation in the place.
His preaching was arresting and people congratulated themselves on having
such a Vicar. He had come straight from France where he had been acting as
Chaplain, and had there shown great bravery under fire. Many came to hear
him just because of this. But when the rage for amusements began to show
itself, and it was found that the Vicar had no sympathy with it, and had no new
Gospel to preach, but preached the same Gospel as they had heard before
the war, untouched with modernism and the various other new religious
theories, the congregation that had increased out of curiosity gradually
dwindled, for they said, "He's not up to date." It was disheartening for Luke,
specially as he heard that a Church not very far off was crowded to
overflowing on account of all the social questions that were discussed during
the sermons, and well-known lecturers on the various religions came down
from London, Sunday after Sunday, to preach.

"Nevertheless," he said one day to his wife, "I shall continue to preach the
Gospel; and by-the-bye Rachel, I must somehow get three days of quiet at
least, for some of the men's Bible Class want to discuss those questions
which have been raised by the Modern Churchmen's Conference; and I must
prepare for the discussion. But I really don't see how I can manage it. I am
late as it is with the Parish Magazine."

Rachel was laying the table for dinner at the time and looked up quickly at her
husband.
"Pass that over to me," she said.

Luke looked at her a little doubtfully.

"Do you think you can manage it?" he asked.

Rachel laughed.

"Certainly I could. I am a little more intelligent than you give me credit for.
Have all the people sent in their accounts?"

"No, that's just it. Sargent has never sent in his description of the Temperance
Meeting, nor has Mrs. Lent of the Scripture Union Meeting. They are so often
late. It means a good long walk as there is no time to send them cards to
remind them of their duty. The manuscript ought to go in early to-morrow
morning to the printer or the magazine will not be out in time. I should be
thankful to have at least two quiet uninterrupted days; but then there are sick
people to visit. I don't see how I can."

"You can quite well if you will only trust me," said Rachel, smoothing the table
cloth. "You have never tried me."

"I wonder what my mother would say."

"What does it matter? I don't belong to your mother I belong to you. You must
take the responsibility of me," she added laughing.

And so it was settled and Luke had two whole days of quiet. He did not move
out of his study except for meals, and then he hurried over them and ate them
without speaking. Rachel, knowing what he was going through and in what
dead earnest he was, in his longing to rid his men of the terrible doubts that
had been sown in their hearts, kept silence. It might have been a quiet day
arranged by the Bishop!

Luke had given her a long list of people to visit, and had told her what
information was needed for the magazine, and Rachel set to work asking no
further questions. She dispensed money where she thought it was needed
(not always wisely alas!) and tended to the best of her knowledge the sick
people, singing to many of them; and though it was work just after her own
heart, being utterly unused to it having had no training whatever, it took a
great deal out of her, particularly as she was of a sympathetic nature. But she
felt it was well worth while when, after the two days were over, the strained
tired look on Luke's face had disappeared giving way to one full of peace and
happiness.

He had felt it his duty to face over again, all the arguments and difficulties that
his people might come across; and he realised that he was at war with the
Devil. The words of St. Paul often ran in his mind; "For we wrestle not against
flesh and blood, but against principalities, against powers, against the rulers of
darkness of this world, against spiritual wickedness in high places."

Not even Rachel was conscious of the spiritual warfare that was raging in her
husband's little study, nor how often he threw himself on his knees crying to
God for help in the conflict, nor had Rachel the faintest suspicion of the
victories that were gained within its walls.

She had often wished that her husband was not so preoccupied, and had
more thought for the small things of life which make all the difference to its
comfort. But though she felt that the things of which she complained in her
heart were so insignificant compared to the great matters about which Luke
was engaged, she had no idea that his preoccupation and absentness of mind
were often caused by the fact that he had either lost or won a spiritual battle.

He was thankful now that he had spent so much time in facing the doubts and
difficulties that he had met with in the course of his reading, so that the two
quiet days enabled him to prepare the subject in such a way as to make the
truth plain to his men.

On the night of the meeting Rachel found it rather hard to occupy her thoughts
with anything but the great strain which she knew Luke was passing through.
He was late home and to turn her mind from that which was making her
anxious she sat down to the piano and sang.

She was just finishing her song when she heard the front door open. She sat
still in her suspense, expecting Luke to come at once into the drawing-room.
But instead, she heard him going upstairs to his study, and walking heavily as
if he was tired.

For some moments she sat still where she was, then she followed him. But at
the study door she stopped.

Was that Luke groaning? Was he ill?

She very softly opened the door and looked in.

Luke was on his knees, his arms on the writing table and his face buried in
them. He was praying out loud.

Rachel closed the door and went downstairs again. Her heart was heavy, and
anxious. She knew that his habit was to pray out loud; but his prayer to-night
was mingled with groans and probably tears. What had happened? Rachel
moved restlessly about the room. Her impulse was to go to the piano, and
soothe her anxiety by playing. But she was afraid of disturbing Luke. Then she
took up her work and sat waiting.

It was late before she heard him coming down the stairs.

His face bore no trace of the anguish he had apparently been going through.
He came and sat down by Rachel's side without speaking.

"Well?" she said.

He was silent for a moment, then he said:

"I have never had such a fight with evil as to-night. I feel sure that the Devil is
working with all his might to destroy any good that may have been done."

"What happened?"

"We have been discussing for nearly three hours the articles of our faith. The
men had primed themselves with all the arguments they could lay hold of
against them. The Divinity of our Lord, the Virgin Birth, the Resurrection; and
the very men who I had hoped were on the eve of making the great decision
have been thrown back."

"But there were surely some who were helped by your words?"

"Yes, thank God. There is a small band of faithful Christians as firm in their
faith as I am. They know the Christ; and believing Him to be God take His
Word as truth. But the greatest number have been shaken by these views that
have been scattered broadcast since the Modern Churchmen's Conference at
Cambridge, and there are some who are weak in the faith and just tottering,
as it were." Luke rose up and began to walk about the room.

"I feel," he said, "that a stronger man than I am is wanted for this place. It
seems to me to be the stronghold of Satan."

"When I am weak, then am I strong," said Rachel.

Luke stood and looked at Rachel for a moment. Then his face broke into a
smile.

"Thank God for my wife," he said.

CHAPTER XVIII.
GAS STOVES VERSUS MOUNTAINS.

"Next year," said Rachel one day, as she and Luke were on their way to
Church, "next year we may not be able to afford a holiday. So I am resolved
that we shall have a good one this August. We will go to the Lakes."

Rachel wrote and secured rooms at Rydal and a month after the men's
meeting mentioned in the last chapter, they started off for the North.

What the sight of the beauty that now surrounded them was to Rachel can be
imagined. She told Luke that she had seen nothing that could be called
beautiful ever since coming to Trowsby, with the exception, she took pains to
add, of her primroses and hyacinths in the little garden of which she was
inordinately proud.

To sit by the Lake in the cool of the evening and watch the lights and shadows
on the mountains, was positive bliss to Rachel. She tried to make Luke revel
in it as much as she did, but alas, his thoughts were still engrossed with his
parish, not withstanding all Rachel's efforts to make him forget it.

"It will be so much better for the parish as well as for you if you will only put it
away from your mind," she said.