Thiele pioneer in statistics Steffen L. Lauritzen all chapter instant download

Download the full version of the ebook at
https://ebookultra.com
Thiele pioneer in statistics Steffen L.

Lauritzen
https://ebookultra.com/download/thiele-pioneer-in-
statistics-steffen-l-lauritzen/
Explore and download more ebook at https://ebookultra.com

Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.
Current Management of the Menopause Christian Lauritzen
https://ebookultra.com/download/current-management-of-the-menopause-
christian-lauritzen/
ebookultra.com
Thinking Politics Perspectives in Ancient Modern and

Postmodern 2nd Edition Leslie Thiele
https://ebookultra.com/download/thinking-politics-perspectives-in-
ancient-modern-and-postmodern-2nd-edition-leslie-thiele/
ebookultra.com
Pioneer Jesuits in Northern Mexico Peter Masten Dunne
https://ebookultra.com/download/pioneer-jesuits-in-northern-mexico-
peter-masten-dunne/
ebookultra.com
Pioneer 1st Edition H.Q. Mitchell
https://ebookultra.com/download/pioneer-1st-edition-h-q-mitchell/
ebookultra.com
Crowds Ethnographic Encounters 1st Edition Megan Steffen
(Editor)
https://ebookultra.com/download/crowds-ethnographic-encounters-1st-
edition-megan-steffen-editor/
ebookultra.com
George Lohmann Pioneer Professional Keith Booth
https://ebookultra.com/download/george-lohmann-pioneer-professional-
keith-booth/
ebookultra.com
Advances in Understanding Strategic Behaviour Game Theory

Experiments and Bounded Rationality Steffen Huck
https://ebookultra.com/download/advances-in-understanding-strategic-
behaviour-game-theory-experiments-and-bounded-rationality-steffen-
huck/
ebookultra.com
Computational Statistics Handbook with MATLAB 2nd Edition

Wendy L. Martinez
https://ebookultra.com/download/computational-statistics-handbook-
with-matlab-2nd-edition-wendy-l-martinez/
ebookultra.com
French Advance and Other Lines 1st Edition Steffen

Pedersen
https://ebookultra.com/download/french-advance-and-other-lines-1st-
edition-steffen-pedersen/
ebookultra.com
Thiele pioneer in statistics Steffen L. Lauritzen Digital
Instant Download
Author(s): Steffen L. Lauritzen
ISBN(s): 9780198509721, 0198509723
Edition: illustrated edition
File Details: PDF, 3.56 MB
Year: 2002
Language: english
Thiele Pioneer in Statistics
This page intentionally left blank
Thiele Pioneer in Statistics
Steffen L. Lauritzen
Professor of Mathematics and Statistics
Aalborg University, Denmark
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Bangkok Buenos Aires Cape Town Chennai
Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata
Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
São Paulo Shanghai Taipei Tokyo Toronto
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© Oxford University Press, 2002
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2002
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose this same condition on any acquirer
A catalogue record for this title is available from the British Library
Library of Congress Cataloging in Publication Data
(Data available)
ISBN 0 19 850972 3
10 9 8 7 6 5 4 3 2 1
Preface
In 1979, on the occasion of the 500th anniversary of the foundation of the University of Copenhagen, Anders Hald
wrote an essay on the history of statistics at the University. Hald's fascination from reading Thiele's books and papers
was contagious, and I too became infected with enthusiasm for Thiele.
In this book I hope to share my fascination with others. For many years I have felt it unreasonable that Thiele's
masterpiece of 1889 had not been published in a language more widely accessible than Danish, and that ‘someone’
ought to put it right and translate it.
In 1997 it became obvious that I was that ‘someone’. Fortunately, I received a grant to stay at the wonderful Institution
of San Cataldo in Southern Italy in the month of September 1998. In such an inspiring, beautiful and secluded place,
miracles are abundant and the first draft translation of the 1889 book was consequently completed.
Back in busy academic life it was difficult to complete the remaining part of the project. Two periods at Klitgaarden,
the former summer residence of King Christian X of Denmark, gave me the necessary peace and quiet. Institutions
such as San Cataldo and Klitgaarden are as important now as they ever were, and for me Thiele will always be
associated with blue sea, olives and lemon trees, as well as brisk winds, rolling waves and blowing sand. And what
experience could be more royal and inspiring than working at the desk of Queen Alexandrine, in the room of Frederik
IX, in the house of Christian X, completing a book about the son of the librarian to Christian VIII, while studying a
painting of him, pictured with Frederik VIII at a meeting at the Royal Danish Academy of Sciences and Letters?
I am indebted to the International Statistical Institute for permission to reproduce material from International Statistical
Review.
In addition to the institutions mentioned, I am grateful to Anthony Edwards and Stephen Stigler for commenting on
the early draft translation. Bjarne Toft deserves sincere thanks for collecting and sending me plenty of valuable
material about and by Thiele, mostly from the files of the late N. E. Nørlund.
My colleagues at Aalborg University should be praised for allowing me to go away and work on Thiele without even
raising an eyebrow.
vi PREFACE
Finally, the existence of this book owes much to Anders Hald, for fascination, encouragement, and discussions, for
allowing me to reproduce some of this work, and also for working carefully through my early draft manuscripts,
supplying an uncountable number of valuable comments at many levels. Indeed, several of the footnotes supplied in
this book stem directly from him.
S.L.L.
Klitgaarden, Skagen
October 2001
Contents
1 Introduction to Thiele 1
S. L. Lauritzen
2 On the application of the method of least squares to some cases, in which a combination
of certain types of inhomogeneous random sources of errors gives these a ‘systematic’ character 7
T. N. Thiele
1 Introduction 9
2 Adjustment of a series of observations of an instrument-constant 13
3 Determination of the units for the two types of weights 25
4 Adjustment by linear functions of the unknown elements 30
5 Numerical example 33
3 Time series analysis in 1880: a discussion of contributions made by T.N. Thiele 40
S. L. Lauritzen
1 Introduction 41
2 The model 42
3 Least squares prediction of the Brownian motion 43
4 Recursive solution to the prediction problem 46
5 Thiele's estimation of the error variances 49
6 Discussion of procedures for estimation of the error variances 50
7 An application 54
8 Final comments 55
4 The general theory of observations: Calculus of probability and the method of least squares 57
T. N. Thiele
1 On the relation of the law of causality to observation 61
1.1 Accidental and systematic errors 62
1.2 Actual and theoretical error laws 63
2 On actual error laws 65
2.1 Actual error laws and relative frequency 65
2.2 Actual error laws expressed by symmetric functions 78
2.3 Actual error laws for functions of observed variables 93
viii CONTENTS
3 The hypothetical or theoretical error law of the method of observation. The law of large numbers 105
3.1 Probability 109
3.2 The error law of the method expressed by symmetric functions 127
3.3 On adjustment 136
3.4 The principle of adjustment by the method of least squares 149
3.5 Adjustment by elements 161
3.6 On systematic errors 179
4 Tables and Figures 185
5 T. N. Thiele's contributions to statistics 198
A. Hald
1 The background 199
2 Thorvald Nicolai Thiele, 1838–1910 201
3 Skew distributions 205
4 Cumulants 207
5 Estimation methods and K statistics 212
6 The linear model with normally distributed errors 214
7 Analysis of variance 222
8 A time series model combining Brownian motion and the linear model with normally distributed errors 223
6 On the halfinvariants in the theory of observations 226
T. N. Thiele
7 The early history of cumulants and the Gram–Charlier series 232
A. Hald
1 Introduction 233
2 The central limit theorem, the moments, and the Gram–Charlier series 234
3 Least squares appromixation by orthogonal polynomials 237
4 Thiele's halfinvariants, their operational properties, and the Gram–Charlier series 241
8 Epilogue 249
S. L. Lauritzen
Bibliography 252
Index 261
CHAPTER ONE Introduction to Thiele
Fig. 1 Photograph of the young T. N. Thiele, reprinted from Nielsen (1965).
Thorvald Nicolai Thiele was a brilliant researcher and worked as an actuary, astronomer, mathematician, and
statistician. This book collects translations into English of his best and most important publications on statistics,
together with some articles written about Thiele which place his work in a modern as well as in a historical context.
Thiele was born in Copenhagen on Christmas Eve, 24 December 1838, and grew up in a prominent family and a
culturally and intellectually stimulating environment. His father, Just Mathias Thiele (1795–1874), was private librarian
to King Christian VIII of Denmark and director of the Royal College of Prints.
2 INTRODUCTION TO THIELE
In 1867 Thiele married Marie Martine Trolle (1841–89). They had six children, four of which survived their first year
(Brenner 1939).
Thiele obtained his master's degree in astronomy from the University of Copenhagen in 1860 and his doctor's degree
in 1866, based on a thesis about the orbits of double stars (Thiele 1866). In 1875 he became a professor of astronomy
and director of the Astronomical Observatory at the University of Copenhagen, positions he kept until his retirement
in 1907. He was the founder and Mathematical Director of the Danish insurance company Hafnia from 1872 until his
death in Copenhagen on 26 September 1910.
There are many articles on Thiele and his statistical work: see for example those by Burrau (1928), Schweder (1980),
Johnson and Kotz (1997), Lauritzen (1999), and Norberg (2001). Thiele was visually impaired by severe astigmatism
and could not carry out observational work as an astronomer (Gyldenkerne and Darnell 1990). He therefore turned
his interests towards mathematical and computational work. In his later years he was almost blind.
Hald (1981) (reprinted in this volume) contains a detailed discussion of his scientific achievements as a statistician,
while more details can be found in Hald (2000a) (also reprinted here) and (Hald 1998, 2000b, 2001). Chapter 14 of
Farebrother (1999) focuses on Thiele's contributions to the method of least squares.
The most famous of Thiele's contributions to actuarial science is his differential equation for the net premium reserve
Vt at time t for a life insurance:
Here π is the premium per time unit, δ is the force of interest, and μt is the force of mortality. In fact he never published
this equation, but showed it to his friend and colleague J. P. Gram, who described it in his obituary on Thiele (Gram
1910). We refer the reader to Hoem (1983) for further discussion of Thiele's contributions in this area.
Other areas to which Thiele made significant contributions include the theory of interpolation (Thiele 1909). Nielsen
(1910) contains an almost complete bibliography of Thiele's mathematical writings. The bibliography at the end of this
book contains items which have been referenced here and in the notes provided to the translations. Some I have
stumbled across and Nielsen has either missed them or classified them as outside the scope of his bibliography on
Danish mathematics.
As an original thinker, Thiele and his ideas were often far reaching and ahead of his time, so much so that he was rarely
understood by his students and contempories. It will appear from the collected texts here that his ideas have
INTRODUCTION TO THIELE 3
tmost importance even today, since they include such things as likelihood, Kalman filtering, the EM algorithm, non-
parametric density estimation (via empirical cumulants), residual analysis, and the method of least squares in which he
was a true virtuoso. That said, he himself may not always have grasped the full generality and scope of his own ideas.
Although he clearly and distinctly expresses the idea of likelihood (page 123), and uses it to discuss estimation of a
binomial proportion, he is not aware that this is a universal idea which extends to other statistical problems. When he
derives the Kalman filter and smoother (Thiele 1880, page 16 of this volume), he is of course not aware of its potential
for on-line computation, and although he describes what is an instance of the EM algorithm (page 25), he has
apparently no idea of its wide applicability. After all, Thiele was also a man of his own time. However, his
understanding of the halfinvariants, introduced in Thiele (1889) (page 82), was complete, or at least became so after
publication of his 1899 paper (Thiele 1899, Chapter 6 of this volume) and predates R. A. Fisher's cumulants (Fisher
1929; Fisher and Wishart 1931). Fisher did not appreciate Thiele's contributions properly and also expressed his
distaste for Thiele's original name: see for example the correspondence between R. A. Fisher and Arne Fisher, R. A.
Fisher and Ragnar Frisch, and R. A. Fisher and Harold Hotelling, the last strongly exposing Fisher's ignorance about
Thiele's work (Bennett 1990, pp. 310–20).
Although few could fully understand and appreciate Thiele's ingenuity, he did have strong interactions with his
colleagues, in particular L. H. F. Oppermann and J. P. Gram, and also a significant effect on the work of his successors,
for example J. F. Steffensen, who wrote the major textbooks used for the degree course on Actuarial Science and
Statistics (established in 1917) and an excellent textbook on statistics (Steffensen 1923a), which may have been the best
one at the time. Steffensen was strongly influenced and inspired by Thiele, but found Thiele's books impossible for use
as textbooks, partly because they were inaccessible to most students, and partly because Thiele had a habit of only
writing original material so his books did not contain the basic material needed to understand the more advanced
concepts (Hald 1983).
Thiele had few students, but he was known as an inspiring teacher although he could sometimes be difficult to follow
(Larsen 1954). The Danish-born actuary Arne Fisher, who emigrated to America at the age of 16, was strongly
influenced by Thiele (Fisher 1922) and took the initiative to have the English version (Thiele 1903) of Thiele's
elementary book (Thiele 1897) reprinted in the Annals of Mathematical Statistics in 1931. A student of Thiele, Kirstine
Smith, went to England to study with Karl Pearson and published two papers in Biometrika
(Smith 1916, 1918), the second of which represents pioneering research in the optimal design of experiments
(Atkinson and Bailey 2001). Kirstine Smith decided to give up life as a researcher and ended her career as a school
master in Denmark (Plackett and Barnard 1990, p. 124).
Having spent a considerable amount of time on Thiele's work, it appears to me that his interests and talents were
strongest in an area which could have been called computational mathematics. When concerned with aspects of
computation, his extraordinary abilities and immediate fascination shine through clearly, even when dealing with very
complex procedures. This impression of mine conforms with the fact that he (unsuccessfully) worked hard to establish
a Professorship of Computational Mathematics (Regnende Matematik) at the University of Copenhagen. Gram (1910)
also emphasized these interests of Thiele as being most prominent and wrote that Thiele was the first in Denmark to
buy and use a calculating machine. Thiele was indeed greatly interested in all aspects of numbers, including their
axiomatization (Thiele 1886). He advocated the use of a number system with base 4 for beginner's arithmetic, base 16
for the advanced, and base 64 for grandmasters (Gram 1910; Larsen 1954). He also had the tiles on the floor of a large
room in the building of the Hafnia insurance company laid according to number-theoretic patterns studied in Thiele
(1874), see Fig. 2 (Larsen 1954). Incidentally, Thiele, with H. G. Zeuthen and H. Valentiner, arranged that Caspar
Wessel's treatise on the geometric interpretation of complex numbers (Wessel 1798, 1999) was translated into French
(Wessel 1897). This work had been ignored by contemporary mathematicians, partly because it was written in Danish
but, perhaps more importantly, because the entire idea of using geometric insight on algebraic problems was
considered unimportant among leading mathematicians at the time (Andersen 1999).
Besides his professional interests, Thiele was engaged in aspects of social policy. In Thiele (1892a) he suggests
consideration of the results of a vote as an outcome of a repeated experiment with fixed probability distribution, and
taking into account the uncertainty due to sampling when evaluating whether the vote is decisive or not. He was
interested in issues of insurance (cf. page 105) and old-age pensions from a social policy point of view, advocating what
today is known as the Schreiber–Samuelsen model for old-age pensions (Petersen 1986; Thiele 1891). Thiele
emphasized that old age is the destiny of us all, as opposed to disability or premature death, which is a random
phenomenon. Therefore a pension system should be seen as a social contract between generations, with the State as
mediator, rather than as an insurance.
INTRODUCTION TO THIELE 5
Fig. 2 Patterns from Thiele (1874). The coloured squares encode complex right-hand sides for which certain algebraic
equations have integer (complex) solutions. The figure should be viewed at some distance.
An important facet of Thiele's personality is represented in his activities as initiator: The Danish Mathematical Society
was founded in 1873 on his initiative in cooperation with H. G. Zeuthen and J. Petersen (Branner 2000; Ramskou
2000). The Danish Actuarial Society was also founded in 1901 on his initiative (Burrau 1926). As mentioned, in 1872
Thiele founded the first Danish private insurance company, Hafnia, where he acted as its Mathematical Director from
1872 and as chairman of its board of managers from 1902 until his death in 1910.
As a matter of curiosity we note that Thiele was an active member of the first Danish chess club, Københavns
Skakforening (founded 1865), from the early days of its existence, and he was prominent enough to be mentioned in
the hymn written by H. E. Gjersing on the occasion of the 50th anniversary of the club (Nielsen 1965).
The core of this book consists of the translations of two of Thiele's most important papers (Thiele 1880, 1899) and his
fundamental book (Thiele 1889). To help the reader appreciate Thiele's work, and to set things in a modern and a
historical context, the translations have been supplemented with comments, also in the form of reprinted papers (Hald
1981, 2000b; Lauritzen 1981), discussing and describing the contents of the translated material. Thiele uses a relatively
modern notation, so this has only been changed in very few places. The few obvious misprints have been corrected
without specific notice and hopefully I have not made too many new ones. When seriously in doubt, I have made a
note. So now it is time for the reader to enjoy Thiele's marvellous texts!
CHAPTER TWO On the application of the method
of least squares to some cases, in which a
combination of certain types of inhomogeneous
random sources of errors gives these a ‘systematic’
character
Presentation
This paper is Thiele's first on the method of least squares. It is a brilliant tour de force, but so far ahead of its time that
only a few could appreciate the results. It sparkles with ingenuity and it initiated my own interest in Thiele's work in
1979. Thiele's recursive algorithm developed in this paper served as an important source of inspiration for Lauritzen
and Spiegelhalter (1988). Enjoy the geometric construction of the Kalman filter on page 26. This may still be a novelty.
Fig. 1 Photograph of T. N. Thiele as a young man. From the archives of the Royal Danish Library.
8 ON QUASI-SYSTEMATIC ERRORS
Thiele must have considered this work important, since he simultaneously had it published in French. Strangely, its
contents are only mentioned sporadically in Thiele (1889), and nowhere else in Thiele's writings. Was he disappointed
because his paper was not properly understood? Or did he not fathom the potential of his own results?
S.L.L.
Fig. 2 The cover page of the original Danish version of Thiele (1880).
ON QUASI-SYSTEMATIC ERRORS 9
Fig. 3 The cover page of the original French version of Thiele (1880).
1 Introduction
Contemporary astronomy is, leaving other important phenomena aside, essentially dominated by its battle against
systematic errors.
The term ‘systematic error’ is used to contrast both with random errors and with errors which can be remedied by
calculating corrections, because the nature and
effect of the sources of error are known. But this contrast is somewhat obscure. There are errors where one can be
uncertain about their true origin long after one has identified the mathematical law describing them as a function of
some external condition of the observations. And the delimitation towards random errors is uncertain because it is
doubtful whether it is possible at all to justify fully that errors can be random. The technical term ‘random error’ is
partly defined by the fact that one can not relate repeated observations to any prior identifiable external condition of the
observations and specify a relation between the errors and such a condition; partly in positive form through
demanding that errors be treated as random so that, when ordered according to size, the number of identical
observations being sufficiently large, they should reveal an error law, identifying the frequency of errors between any
arbitrarily chosen limits as a function of these limits.
In the definition of random errors some would perhaps further demand that this error law should take a specific
functional form, the exponential1
where x0 is the most likely value and m is the standard deviation. But even though one would not consider errors as
being systematic when the error law is different from the exponential, which is a prerequisite for direct application of
the method of least squares, there are cases where the boundary between the two types of error can seem blurred.
In several areas, in particular in connection with the so-called personal errors which seem to have some physiological
origin, and in connection with series of observations made to determine specific aspects of the settings or functioning
of an instrument, to be denoted instrument-constants, as they would be constant had the instrument been perfect,
cases occur where it is all too obviously impossible to characterize the errors by an error law and where, when
observations are arranged chronologically, there seems to be a dependence between the time of observation and the
size of error, so at least initially it would be justified to classify the errors as systematic. But when it has been attempted
to correct such errors under the assumption that they were mathematical functions of the time of observation, the
hope of ridding the observations of these embarrassing systematic errors has been perpetually frustrated. When
believing that the appropriate correction formula
1
At the time, ‘the exponential error law’ was the standard name for the Gaussian or normal distribution, tr.
has been found, and trying it out on sufficiently prolonged series of observations, it has persistently appeared
impossible to represent the errors as functions of time or at least to use the formula for prediction of personal
observational errors or instrument-constants.
It could be that more satisfactory results would be obtained in such cases through a more detailed investigation of the
relation of these errors to random errors. If the errors are not directly random, it is at least possible that they could
result from the combined effect of several sources of random error, because the well-known theorem that a
combination of several sources of random error, each with its error law, can be described as a single source of random
error, with its own error law depending on the error laws of the others, has an exception which does not always seem
to have attracted sufficient attention. The theorem does not hold when some of the sources of error act in such a way
that more than one of the otherwise independent observations are affected by every single error flowing from the
source in question. Let us for example assume that a series of values o1, o2, o3, …, on that themselves could have been the
proper object of observation, have been influenced by a random source of error which has displaced these by e′1, e′2, …,
e′n, according to an error law which can be assumed exponential, and further that one is not considering the o as objects
of observation, but their successive sums arranged according to their subscripts, i.e.
and that one in addition must consider the problem in this fashion because other significant sources of random error
are affecting the sums with random errors e″1, e″2, …, e″n having a different error law which could also be assumed
exponential, then the totality of errors (1.1)
will in general follow neither the exponential nor any other error law but will, when arranged after subscripts of e″,
display phenomena very similar to what is seen
for chronological series of personal errors or instrument-constants as mentioned above. There are also rather good
internal reasons to explain these phenomena in this or a similar fashion through a combination of random sources of
error, and in this respect it is about the same whether it be the personal errors or the errors of instrument-constants, as
the sensory organs could be conceived of as instruments. The errors would partly be such errors which would vary at
random from one observation to the next as the errors e″ in our example, partly such errors which necessarily are
determined by the random state of the instrument at the moment of observation, to the extent that this cannot be
calculated from conditions such as temperature, pressure, etc.
As regards the state of the instrument one should not, when the same instrument is used, ignore a dependence on the
state of the preceding moment. Certainly it is true for many instruments that there is a normal state from which the
instrument will never deviate more than what makes it stay within certain limits, and then it will most likely often have
a higher probability of drifting towards this normal state than of a further deviation. Still I have a strong inclination to
believe that such random variations of an instrument, making it fluctuate around a normal state as the most likely state
at all times (making these errors combine with the errors of observation e″), would be rather rare, and at least not at all
as common as variations where the influence of the normal state is negligible, and where the ‘status quo’ is the rule, so
the random variations of the instrument are realized in such a way that the state at any given moment is the most likely
state at the next moment, having as its consequence that the errors accumulate like e′.2 A third type of instrument
variations should be mentioned in this context, namely variations where the actual state at any given moment is equal
to the average of the state in the preceding moment and the most likely state in the next; such a relationship is difficult
to imagine for instrument-constants, but for the clock it should play a dominant role and make the errors in the
movements of the clock act through a double summation.3 I have attempted to develop the theory of errors which are
generated by two sources of error as in (1.1) and hope thereby to have made a contribution towards reducing the
domain of systematic errors. However, there would be a considerably greater benefit by learning how to calculate the
corrections
2
This argument, together with the formula (1.1), essentially derives what is today known as the model for the Kalman filter, consisting of Brownian motion with additive
independent noise. Clearly Thiele does not have the mathematical concept of a stochastic process at his disposal, but the idea is very clearly expressed, tr.
3
And here integrated Brownian is motion suggested as a model of clock errors! tr.
which fully compensate the systematic errors than by referring them to the domain of random errors and make way
for application of the method of least squares in cases where it hitherto was inapplicable; but not every evil can be
pulled up by the roots and it is conceivable that occasionally one will deal with errors where the condition for revealing
the acting sources of error and their laws exactly is to partition these into two: those which independently act on
individual observations, and those which act through cumulation.
That this is possible follows from the fact that there already exists a solution of the problem in a special case.4 If such a
series of observations is partitioned into groups separated by long time intervals and if the observations in each group
are made so quickly after one another that they can be considered simultaneous, then the problem is not difficult; the
mean of observations within each group is calculated to yield the most likely value for the group, together with the
standard deviations of a single observation and corresponding standard errors of the means; in addition one is
gathering information about the most likely values in the intervals between groups of observations by interpolation,
assuming a steady change from one group to the next; and the differences occurring in the long time intervals can be
exploited to judge the stability of the instrument. Then the problem should also be solvable when the intervals are
filled with scattered observations and where the number of repeated observations within groups is reduced. The only
difference should be that it becomes somewhat more difficult to derive the desired results.
The principle leading to a solution of these problems is not at all sophisticated. When formulating the problem in
terms of equations, one should include every equation to give a complete description of the assumed relationships, and
not permit any kind of elimination which is not prescribed by the method of least squares. Perhaps it should be noted
here that the analytic tool used to implement the method, a continued fraction, is somewhat unusual in practical
astronomical calculations, but is in no way inconvenient.
2 Adjustment of a series of observations of an instrument-constant

We assume an instrument to have a property which is expressed by the number x and which should have been
constant, but in reality x is subject to changes, either continually or at known moments in time. We assume in addition
that each of these
4
Thiele refers to the method of normal places, described for example in Thiele (1903, pp. 106–12); see also Farebrother (1999, pp. 214–16) and Hald (1998, p. 393). tr.
changes is completely random and follows an exponential error law determined by the current value of x being the
most likely value of x at the next moment, whereas the mean square errors of x or rather of its momentaneous
changes, μ, axe known. If we then let x0, x1, …, xn denote the values taken by x at a series of times t0, t1, …, tn (either ta + 1
always later than ta or always earlier than ta), at which either observations of x are made or at which one may wish to
know x, then by assumption 0 will be the most likely value of every difference xa + 1 − xa; the mean square error of this
difference will be (between the limits a and a + 1), and the absolute weights5 are thus
assumed known. In this way we obtain n equations (2.1)
We further assume that observations of x made at times t0, t1, …, tn also are subject to random errors with exponential
error laws such that for these observations z0, z1, …, zn we know the most likely values (before adjustment) and also
know their mean square errors m0, m1, …, mn or their weights ν0, ν1, …, νn where in general νa = 1/ma. This gives n + 1
equations (2.2)
which together with (2.1) contain all the mutually independent conditions we have available to determine the n + 1
unknowns x0, x1, …, xn which are sought together with their mean square errors after adjustment.
Note. To the extent that among the times t there are some where no real observation has been made, and at which we
simply wish to know the most likely values of x, we should let the corresponding weights be equal to ν = 0, whereafter
the values of z can be arbitrary (not infinite).6
5
Thiele has here identified the basic properties of Brownian motion as having increments which are independent and with variance proportional to the length of this time
interval. tr.
6
This idea is in Thiele (1889) further developed into a general method of fictitious observations, an alternative to modern techniques for solving least squares problems using
generalized inverse matrices, cf. page 171 of this volume. See also Hald (1981) and Hald (1998, p. 648). tr.
If we apply the method of least squares to (2.1) and (2.2) we obtain the n + 1 equations (2.3)
Among these equations we should first eliminate those xa which have νa = 0, that is corresponding to fictitious
observations which are only included to determine the values of x at other than proper observation times. But νa = 0
implies (2.4)
so, when first xa − 1 and xa + 1 have been found, xa is determined by simple interpolation between these values under the
assumption that the changes in x are inversely proportional to the weights between the limits of the interval in
question, or directly proportional to the mean square errors, or with the length of the time interval itself in so far as the
prospects of displacement of the instrument have been constant over the time interval. If the first or last observation z0
or zn was ficticious, that is if the values of the instrument-constant before or after the period of observation are sought,
then ν0 = 0 implies that x0 = x1 and νn = 0 that xn = xn − 1so the x obtained for the first and last regular observation times
are the most likely values for the past and future respectively.
The equations (2.3) thus show what is also completely obvious: that the elimination of xa when νMa = 0 does not change
these equations in any other way than yielding the result which one would directly have obtained by omitting the
equations concerning the fictitious observation. One should then only let
or
and proceed directly from the equation
to
omitting the equation for νaza. When performing these calculations in real examples we could therefore directly assume
that no fictitious observation is included, knowing now how to derive x at an arbitrary time from the x already found
and corresponding to real observation times.
Naturally the equations (2.3) must be further treated in the way typical of the method of least squares, but here this is
made easier by the comparatively simple form of the equations. To take advantage of this and to represent the solution
in the most suitable way, auxiliary variables must be introduced. Firstly, instead of the weights νa and νa, a + 1 we introduce
two other series of numbers determined by the recursion formulae:7(2.5)
and (2.6)
Under the assumption that u0 = 0, ua and ua, a + 1 are easily calculated using a table of reciprocals. (No greater accuracy is
needed beyond what has real significance for the weights.)
7
Thiele is now deriving the recursive algorithm known today as the Kalman filter and smoother (Kalman and Bucy 1961): first a recursion for the weights, then a
forwardpass for filtering, and finally a reverse pass for smoothing. tr.
Next a sequence of values y0, y1, …, yn are calculated from the z which in the further calculation will replace the z and
which in addition will turn out to represent the corresponding value of x under the assumption that all observations
with higher subscripts are ignored; thus we have at least yn = zn. Such y are found by the recursion formula (for
increasing subscripts)8(2.7)
or more conveniently (2.8)
or (2.9)
in combination with y0 = z0.

Substituting these into (2.3) one finds that the x are determined by the recursion formula (for decreasing
subscripts)9(2.10)
or with more convenient calculation (2.11)
or (2.12)
in combination with xn = yn.

That this is not just correct, but that (2.10) is indeed also the typical form of elimination according to the method of
least squares, is proved by general induction. For a = 0 (2.10) coincides with the equation (2.3), and if we assume that
(2.10) holds for all subscripts up to a, that is the elimination of x0, x1, …, xa − 1 has left a system of n − a + 1 equations,
namely (2.10) and the n − a last equations
8
Here comes the forward filter. tr.
9
And this is the backward smoother. tr.
of (2.3), then from the two first of these equations
being the only equations in which xa is still occurring, the elimination of xa is (in the usual way) made through
multiplying the first equation by
cf. (2.6), and adding the result to the second of these equations, whereby from (2.6) we obtain
and thus according to (2.7) and (2.5)
which coincides with (2.10) and thus proves the correctness of the statement when a < n. Next, it is easily verified that
the last elimination leads to unyn = unxn.
If an explicit expression for un is expanded according to (2.5) and (2.6) one obtains the continued fraction10(2.13)
and if the first couple of terms are ignored, the final continued fractions will alternate between ua and 1/ua, a + 1.
10
This continued fraction representation of the recursion for the weights has seemingly not been given elsewhere. Thiele had a strong interest in continued fractions (Thiele
1869, 1870, 1879). tr.
From the importance this continued fraction has to our problem it can be assured that simpler solutions will occur
only when this continued fraction takes on a simpler form, and thus that approximative solutions with easier
calculations are only possible in cases which are similar to these. One would first think of cases where the continued
fraction was periodic, but this does not yield simplifications of the practical calculations. This is, however, the case
when the continued fraction has been interrupted because a partial denominator becomes infinite, that is either νa = ∞
or νa, a + 1 = 0, and in the case where the continued fraction can be combined into fewer terms because a partial
denominator vanishes, that is either νa = 0 or νa, a + 1 = ∞.
When either an observation can be considered error free, νa = ∞, or the observational series is interrupted by a long
time interval or by an interval in which it is feared that the instrument has been disturbed, νa, a + 1 = 0, then in both cases
the continued fraction is interrupted or partitioned into two mutually independent continued fractions. This in turn
partitions the computation of the x in the first case such that ua, a + 1 = νa, a + 1 and za = ya = xa, that is such that the two
series are connected through the common value xa = za; in the second case such that ua + 1 = νa + 1 and xa = ya but ya + 1 =
za + 1 so the parts become completely separated by at least one intermediate value of x which cannot be determined.
Such partitionings would in many other cases imply great advantages to the practical calculations; here the benefit is
not so great for one will almost always be able to partition the calculations arbitrarily without much disadvantage, other
than a minor double calculation of the u, y, and x around the point of partitioning, which quickly will tie the
calculations together so the new values turn out to be equal to those found through the partitioning. The present
consideration just shows that, when there is a choice, one should partition such that a reliable value of z is included in
both pieces and such that one is interrupting at a place where there is a large time interval between two consecutive
observations of which the first terminates the first part and the second initiates the second part.
For the cases in which the number of terms in the continued fraction can be reduced, it is enough to consider
νa, a + 1 = ∞, as we have already considered the case ν = 0 from another point of view and seen that if the problem is to
find x at a time where no observation has been made (or only observations so unreliable that their weights should be
set to be vanishingly small), the most likely values should be calculated by simple interpolation between the closest
previous and the closest subsequent time of observation, under the assumption of a steadily changing instrument. On
the other hand, if νa, a + 1 = ∞, which is implied by the times of observation being so close to each other that the
instrument cannot have changed
its state in the meantime, then it is obviously justifiable to contract consecutive observations to their appropriately
weighted averages and thus consider
as a single observation instead of za and za+1; this is also confirmed by our theory which for νa, a+1 = ∞ gives
and
When I previously have been careful not to deviate from the typical process of elimination by the method of least
squares, it was to take advantage of the important theorem which I only know by oral communication from Professor
Oppermann (Science Meeting, Copenhagen 1873) that there exist systems of linear functions11 of observations, which
in complete generality could be considered as mutually independent observations and as such could substitute the
original observations, in particular during all calculations of mean square errors of the elements in the adjustment; and
that the left-hand sides of the equations resulting from the process of elimination, here (2.10), represent such a system
of functions that could be considered as independent observations.
Thus in
11
In Thiele (1889), cf. page 144 of this volume, these are studied in detail and named free functions. They correspond to orthogonal linear combinations in modern
terminology. tr.
the products u0y0, …, uaya, …, unyn can be considered as independently observed values such that the mean square error
of uaya is equal to
and consequently the values of y could themselves be considered independent so that their mean square errors become
(2.14)
with the single exception that (2.15)
If we now express the x explicitly as linear functions of the y it becomes very easy to express the mean square error of
any result from the adjustment which in advance has been given as a linear function of the x. But as
the coefficients in the equations (2.16)
will be given as (2.17)
One then obtains for the mean square errors

and in general
From the first of these expressions one derives the recursion formula (for decreasing subscripts) (2.18)
We note in particular the special case of the general formula which gives the mean square error of a linear function of
two consecutive x:
because xa+1, being a function of y of higher subscript only, can be considered an independent observation besides ya.
Setting r = 1, s = −1 yields (2.19)
The equations (2.18) and (2.19) serve conveniently to calculate the mean square errors which are left after adjustment
in the equations (2.1) and (2.2) emerging directly from the basic conditions of the problem. From these the more
general equation for M2(rxa + sxa+1) can also be transformed to the following:
Recalling that, according to (2.4), the x with no simultaneous proper observation can be calculated from the x
corresponding to the closest preceding and subsequent observation as a linear function of these, say rxa+sxa+1 with r+s
= 1, it would seem obvious to use this equation for M2(rxa + sxa+1) to calculate the mean square error for the x which
are not determined by simultaneous observation. However, this would be a serious mistake of a special type of which
one should beware in similar investigations. Because although xb = rxa + sxa+1, one should not let M2(xb) = M2(rxa +
sxa+1) in cases where the equation xb = rxa + sxa+1 itself is a result of adjustment and therefore has no absolute validity,
but only is an approximation with a reliability which could be measured by a mean square error of M2(xb) − M2(rxa +
sxa+1).12M2(xb) must be determined directly in the same
12
Thiele must give this rather complex explanation because he does not use different symbols for the true value x and its estimate x̂ In modern symbolism he warns against
using the equation x̂b = rx̂a + sx̂a+1 , as if it were true that xb = rxa + sxa+l . tr.
way as xb, namely by the insertion of a fictitious observation. The consequence of such an insertion is probably best
explained through changing the original scheme of the ν, u, and x
to the following
according to the relations
together with
where thus
whereas
As regards the y as independent observations, the only difference is thus that the original ya, having weight ua − ua,a+1, has
been replaced with two, namely ya
with weight ua − ub and yb = ya with weight ub − ua, a+1, such that their values are identical and the sum of the weights
equals the original weight. However small this difference would seem, it has genuine significance, but only in regard to
the determination of M2(xb). From (2.18) one now obtains directly
double application of (2.18) further gives
yielding, as ub = ua,b,
which is unchanged from (2.18). One thus has
and (2.20)
consequently the result is that
If in these formulae we let va, a+1 = 0, the problem becomes to determine the x and their mean square error within a
time interval embraced by the observations, or in an interval where the instrument may have been subjected to such a
disturbance that the series of observations must be considered completely interrupted, then, according to (2.6), νa, a+1 =
0 implies that ua,a+1 = 0. Unless νa,b = 0 also, it
holds that p = 1, xb = xa, and M2(xb) = 1/ub = 1/ua + 1/νa, b which in itself is natural.
Figure 4 on p. 26 shows a geometric construction of the adjustment.13
3 Determination of the units for the two types of weights

When any adjustment has been made, the question perpetually arises whether the errors left in the observations satisfy
the assumptions upon which the adjustment rests; if one cannot or will not test this, the result must be considered
hypothetical. Among other things one should attempt to answer the question whether the assumptions about mean
square errors or weights on which adjustment by the method of least squares is based agree with what a posteriori can
be derived from the table of residuals. Often this question takes the special form that one a priori does not know the
weights of the single equations but only some ratios between them, which would be sufficient to determine the weights
if in addition one knew a small number of units for the weights, one for each system. Then one must include these
units among the unknowns of the problem and because of the complex nature of the problems, one is obliged to use
the indirect method, beginning with completely hypothetical units for the weights, and calculate new and generally
improved values for the units, based on the residuals after completion of the adjustment, whereafter the adjustment is
calculated afresh until the posterior units from the last adjustment agree sufficiently accurately with those calculated
prior to the same adjustment.14 Such recalculation is only unnecessary in one particular case, namely when only a single
common unit for all the observations is unknown. In this case the absolute value of the weight unit has no influence on
the adjustment and is itself determined by the formula
where f is the residual in a single equation, n the number of equations, and m the number of elements, or possibly by
13
Thiele's original text has no reference to the figure so this sentence has been added by me. It is not clear why Thiele produces this diagram. Most likely he is simply enthused
by the idea, as I am. To the best of my knowledge this geometric version of the Kalman filter and smoother has not otherwise been described in the literature. tr.
14
The procedure suggested, iterating eqn (3.1), is in effect an instance of the EM algorithm, cf. Lauritzen (1981). tr.
Fig. 4 Geometric construction of the preceding adjustment

(1) The points ζ0, ζ1, …, ζn are marked on an axis of abscissae in order of subscript such that everywhere ζa + 1 − ζa = 1/
νa, a + 1, thus representing the times of observation if the mean displacement of the instrument has been constant over
time intervals of equal length. (2) After addition of a common constant to make all observations positive, the points za
are marked vertically above ζa with ordinate za − ζa = observation za. (3) Points υa are marked vertically below ζa with
ordinates ζa − υa = 1/νa. (4) On the same ordinates, a series of points φ1, …, φn are marked vertically below υ1, …, υn and
on the axis of abscissae a corresponding series of points η0, η1, …, ηn, all of them before the ζ with identical subscript.
The construction is made sequentially by increasing subscript and begins with η0 being determined by ζ0 − η0 = ζ0 − υ0
and subsequently everywhere such that υa − φa = ζa − ηa − 1 and (ηaυa) ‖ (ηa − 1φa)15 (5) The points y0, y1, …, yn are now
marked vertically above the η with identical subscript and situated on a system of straight lines, each ya on the straight
line between ya − 1 and za, in particular y0 with an ordinate of the same size as z0, y0 − η0 = z0 − ζ0. (6) The points x0, …, xn
representing the adjusted values are then determined from they such that every xa is situated on the same ordinate as
the corresponding za (and ζa), xn − ζn = yn − xn, and subsequently xa on the straight line between xa + 1 and ya whereby the
line segment between xa and xa + 1 represents the most likely movement of the instrument in the time interval between ta
and ta + 1.
15
Thiele uses the symbol ≠ for parallel rather than ║. tr.
to the extent one should also include the weight unit among the unknowns on the problem, which, however, must be
considered rather problematic. Examples of cases with more than one unknown weight unit are very common in
astronomy: one just needs to think of the weights of the right ascension and declination under orbit determination.
Examples of a correct treatment of cases with several unknown weight units are not common.
In the special problems treated here it will most often be necessary to repeat the adjustment at least once with
improved weights, because, while one would frequently know the ratios between each of
between each of
separately, one would far from always know the units for the weights νa with sufficient accuracy and most often be in
an even more unfavourable position concerning the weights νa, a + 1. And notably, because of a fact to be mentioned
immediately, it would be unavoidable to repeat the adjustment, although there is reason to believe that the convergence
in the determination of the units for the weights most often will be so rapid that the work would not easily become
prohibitive.
When an adjustment has been calculated on the basis of a hypothetical unit Eν for the weights ν0, ν1, …, νn and another,
Ew, for the group ν0, 1, ν1, 2, …, νn − 1, n one should clearly attempt to determine Ev from the differences za − xa and Ew from
the differences xa − xa + 1. There are n + 1 differences of the first type, n of the second type, and n + 1 elements x0, x1,
…, xn have been determined by the 2n + 1 equations. Possibly one should add 2 to the number of elements (i.e. for Eν
and Ew), but by the large values which n will always have in this type of investigations, this question becomes of quite
secondary importance and I shall therefore ignore this possibility in the sequel and stick to n + 1 as the number of
elements. For the desired units one can then write
where at this point we only know about ξ and η that their sum is equal to the number of elements
and we must determine what fraction of this number to allocate to each of the groups. In other problems, where the
number of equations is large compared with the number of elements, this does not matter much and it will often be
sufficient to distribute the number of elements evenly. But here the number of equations is never quite double the
number of elements and, in addition, an example will easily show that the distribution sometimes should be very
uneven. For if we imagine that an instrument-constant was only effectively observed at two different times, but each
time with with numerous repeated observations, one immediately after the other with no possibility of real change in
the instrument, then it is easily seen that η would approach the limit n − 1 while ξ approaches the value 2. To be able to
determine ξ and η in more complex cases and in general it must be investigated in detail how much each equation
contributes to the number ξ or η which should be subtracted from the number of equations, or how much one should
add to νa(za − xa)2 or νa, a + 1(xa − xa + 1)2 to be able to assume that the resulting sum on average should be equal to 1 for
every equation. In the last formulation the question has an easy answer: because after adjustment xa and xa − xa + 1 are
subject to uncertainty with mean square errors M2(xa) and M2(xa − xa + 1) respectively, see (2.18) and (2.19). Since this
uncertainty, through the requirement that the sum of squared deviations times the weights of the equations should be
driven to its minimum, is in general causing the sum of squares to be smaller than if the observations were compared
with the true values, it is at least very plausible that the right value is found by adding the mean square error of the
residual to each of the squared deviations. A similar procedure is used to show succinctly that the number of
observations should be reduced with the number of elements in the formula for posterior determination of the
weights. In our case it is shown without too much difficulty from (2.16) and (2.17) that
For the two groups of equations we find in this fashion16
and noting that these values of ξ and η) are completely independent of any common factor of Ev and Ew and usually
rather stable under changes of the ratio Ew/Ev, we
16
The quantities ξ and η are what Wahba (1983) named the ‘equivalent degrees of freedom’ for error and signal respectively. In the numerical example of the paper, Thiele
uses the expression ‘excess equations’, see page 34. tr.
obtain for the determination of the units of weights (3.1)
That we have made the last adjustment correctly including the right-hand sides of (3.1) with correct weights is thus
indicated by Eν = 1 and Ew = 1.
At this point we note that while M2(xa) cannot be simplified advantageously by (2.5) and (2.18), we have according to
(2.6) and (2.19) that
which makes us realize that when νa, a + 1 is relatively large and consequently ua − ua, a + 1 small, νa, a + 1M2(xa − xa + 1) will
typically only be a little less than 1, so in cases where the observational times ta and ta + 1 are only a little different from
each other, the differences xa − xa + 1 will almost not contribute to the determination of Ew. This implies that in such
cases the adjusted values xa and xa + 1 will not easily be much different and from this we infer the general rule that
before an adjustment of this kind, one may and should contract all groups of observations which are close in time to
their averages; with a little care one is not going too far with such a contraction, which obviously yields a considerable
reduction of work. Of course one should not miss the chance to use the observations contracted to their averages in
the process of determining the unit of weights Eν. In this way one would frequently obtain such a good determination
of Eν that final adjustment will only modify it slightly. It is more difficult in advance to ensure a sufficient approximate
determination of Ew. In this respect I recommend making a rough calculation after the principle that one first contracts
consecutive observations to their means as far as possible and subsequently considers these as adjusted values for x,
while totally ignoring scattered observations which could not be used in this process, thus obtaining a series of values
Correspondingly one must contract the weights νa, a + 1 tentatively specified with an arbitrary unit, such that
etc. If then the unit of weights Ew is calculated as
where N + 1 is the number of means xA used, then such a determination will tend to be somewhat too large but still
usable as a first approximation.
But even when one has confirmed that the weights used in an adjustment are correct under the assumption that the
error laws are exponential, there is still a need for much more criticism than usually applied. ‘Skew’ error laws and
error laws with multiple maxima may rarely appear, but demand a different treatment when they do. It is not easy to
get rid of systematic errors.
4 Adjustment by linear functions of the unknown elements

When an adjustment of a series of observations subject to a combination of two sources of random errors with
different effects has been completed according to the rules laid out here, one must naturally still further investigate the
two tables of errors. It could be that the results show deficiencies which indicate that the combination has been of a
different kind than the relatively simple one treated here. It could also be the case that, after distinguishing the two
types of error, signs of proper systematic errors are revealed in one or both of the tables of residuals. These errors
might have been hidden by the combination between the regular errors of observation and the displacements of the
instrument, but after this complication has been resolved, they may show so unmistakably that there is hope to remove
them by calculating corrections based on some side condition of the observations. In addition we have until now only
considered the case where the observations subjected to complicated errors relate to a phenomenon which itself could
be assumed constant. It should thus still be shown briefly what to do when the observed phenomenon varies
according to laws which could be assumed to be linear functions with a number of unknown coefficients and
simultaneously is subject to error of the complicated type treated here. It suffices here to treat the case where the law
for the observed phenomenon is ζa = pfa + qga where ζa, the true value for the observation made at time ta, depends on
two unknown elements p and q, while fa and ga are known numbers depending on ta. The observations are assumed
subject to two types of error: partly such xa (corresponding to ta) which owing to the changing state of the instrument
would be the same for every observation made with this instrument at time ta, but vary at random from one
moment to the next so that one has equations (4.1)
with corresponding weights νa,a + 1 as above; and partly such errors which in a random and mutually independent way
affect the observations such that the combined action of the two yields za rather than ζ a as observed values, that is
equations (4.2)
with weight νa, corresponding to the equations (2.2) in the simpler problem. With n + 1 observations (indexed from 0
to n) we thus have 2n + 1 equations to determine n + 3 unknowns x0, x1, …, xn, p, and q. By the method of least
squares we obtain17(4.3)
where generally [νrs] = ν0r0s0 + ν1r1s1 + ⋯ + νnrnsn. When eliminating these equations we still use ua and ua,a + 1 given by
(2.5) and (2.6), that is the continued fraction (2.13) also plays a major role here. Also we use the same auxiliary
variables as denoted by y above although here we use another symbol for these, namely {z}, that is {z0} = z0 and
generally
and further two sets of values, {f} and {g}, calculated in an analogous way from f and g: {f0} = f0, {g0} = g0, and
generally
17
Compared with Thiele's original an apparently redundant ⋯ + between νa, x
a + 1 a
+ and νafap has been removed in this formula, tr.
and in addition some more complex albeit analogous auxiliary variables
all five of them being calculated by the same general recursion formula
beginning from {r0s0} = r0s0.

Using these auxiliary variables the first n + 1 of the final equations become (4.4)
leading to two equations for determining p and q(4.5)
analogous to those one would have found in the usual way if za = fap + qap had been subject only to ordinary random
errors, that is
Here one can derive simplified versions of the recursion formulae for the auxiliary variables. For {ra + 1} one obtains in
analogy with (2.8) and (2.9)(4.6)
or (4.7)
and for {ra + 1sa + 1} we have

or18
which by introducing yet another series of auxiliary variables can be used to obtain formulae which are adequate for
practical calculations so that (RS)a is given by (4.8)
while using this to calculate {ra + 1sa + 1} as (4.9)
We finally note the formula
which can be of significance as a control of (4.8) and (4.9) as well as (4.6) and (4.7).
After having derived the basic formulae for this problem, I should mention that when fa is constant and equal to f for
all subscripts, it also holds that {fa} = f and {fara} = f{ra), which implies that the constant term fp combines inseparably
with the x so that xa + fp replaces xa in the equations (4.4) and the penultimate term vanishes, while the first of the
equations in (4.5) remains unchanged and the second simplifies. Thus, in so far as the equation for the observed
phenomenon contains a constant term, this cannot be separated from the errors x. Consequently the given solution
applies to equations of the form ζa = 0 + pfa + qga without changes.
5 Numerical example
It would probably not be superfluous to illustrate the simplest case of this theory with a numerical example. For this
purpose I choose a series of observations which have been taken during the determination of longitude between Lund
and Copenhagen this summer to determine the index error of the micrometer on the meridian circle of Copenhagen. I
have here used the same weight for all observations of the coincidence between a movable thread and a fixed thread,
and I further
18
Two terms have been moved from the left-hand side to the right-hand side of this equation compared with Thiele's original, tr.
assume that the point of coincidence between the two threads, which appears not to have been constant, has varied in
a completely random way from one moment to the next, and also that the mean change of the point of coincidence has
been the same in each element of time, thus that νa, a + 1 is inversely proportional to ta + 1 − ta whether the micrometer has
been used in the meantime or not; however, it will become apparent that I have attempted to test the last assumption
by a special investigation. Of course I have assumed that both error laws are exponential and free of systematic errors,
although it will turn out to be doubtful whether the latter has been quite justified.
The observations are represented in units of of a revolution = 0.2633″ and to save space I throw away what is
common to all observations of tens, hundreds, etc., of this unit. In accordance with this the unit of weights Ev and Ew
are determined such that Eν = 1 would mean that the standard deviation of a single coincidence was 0.2633″ while Ew
= 1 would demand that the standard deviation of the displacement of an instrument in a single day would be 0.2633″.
From the observations contracted to averages, Eν was tentatively determined from 36 mutually independent equations
each with weight of a single coincidence having its sum of squared deviations equal to 2.902, yielding Eν = 12.4.
From a tentative adjustment, based on assumptions which partly proved to be incorrect, I had found Eν = 12.5 and Ew
= 24.8. These weights were used to calculate the values for νa and νa, a + 1 which together with za form the basis for the
following calculation. To ease the calculations I have in reality used the double weights everywhere, but for clarity I
have reported the results of the calculations as if this were not done. An accuracy of two digits was sufficient.
The fact that the sum ∑νaM2(xa) + ∑νa, a + 1M2(xa − xa + 1) is not exactly equal to 74 is essentially due to the uncertain
determination of the terms in parentheses in the second part of the sum, corresponding to very small time intervals.
To the determination of we thus have a sum of squares of 53.22 in 74 - 17.77 = 56.23 excess equations19 and
further a sum of squares of 12.5 × 2.902 = 36.27 in 36 equations from the observations contracted to their average;
that is, in total 92.23 equations with a sum of squares equal to 89.46, whereby = 1.03. For we have 73 - 56.23 =
16.77 excess equations and a sum of squares of 16.73, that is = 1.00. The agreement is thus good enough; and thus
the result is that the absolute weight on the observation of a single coincidence is equal to 12.9 corresponding to a
standard deviation of ±0.278 = ±0.073″, while the stability of the micrometer in one day of stellar time is measured
with the absolute
19
These are the ‘equivalent degrees of freedom’, see page 28. tr.
weight of 23.8 corresponding to a standard deviation in the same time period of ±0.205 = ±0.054″. Concerning our
assumption that the weights simply should have been proportional to the length of the time intervals, it is seen from
the following more specified table
Interval excess equations sum of squares E′w
less than 0.5 day 3.69 4.91 0.75
between 0.5 and 1 day 7.08 6.15 1.15
more than 1 day 6.00 5.67 1.06
that it seems that the weights should have been set at somewhat smaller values in the short time intervals where the
micrometer has been heavily used, but this result is so uncertain because of the modest number of equations that it is
not at all necessary to modify the assumption that the extent of usage of the micrometer
Index a 1879 Juli va ua ua,a+1 M2(xa) M2(xa - xa + vaM2(xa) vM2(xa-x-
1
) a+1
a, a+1
0 2.61 25.0 25.0 0.0180 0.45
0.0100 20.0 0.0086 0.86
1 2.85 25.0 45.0 .0156 .39
.0336 18.0 .0182 .54
2 3.65 25.0 43.0 .0132 .33
.0088 31.5 .0070 .80
3 3.86 25.0 56.5 .0126 .32
.0316 20.5 .0156 .49
4 4.61 37.5 58.0 .0108 .41
.0016 53.0 .0014 (.88)
5 4.65 25.0 78.0 .0114 .29
.0832 10.5 .0214 .26
6 6.63 25.0 35.5 .0136 .34
.0004 35.0 .0004 (1.00)
7 6.64 25.0 60.0 .0134 .33
.0420 17.0 .0214 .51
8 7.64 25.0 42.0 .0182 .46
.0412 15.5 .0242 .59
9 8.62 12.5 28.0 .0232 .29
.0420 13.0 .0246 .59
10 9.62 25.0 38.0 .0186 .46
.0436 14.5 .0222 .51
11 10.66 25.0 39.5 .0150 .38
.0100 28.0 .0086 .86
12 10.90 25.0 53.0 .0150 .37
.0626 12.0 .0198 .32
13 12.39 25.0 37.0 .0092 .23
.0012 35.5 .0010 (.83)
14 12.42 25.0 60.5 .0090 .22
.0008 57.5 .0008 (100)
15 12.44 25.0 82.5 .0090 .23

.0110 43.0 .0086 .78
5.50 10.82
16 12.70 25.0 68.0 0.0120 0.30
0.0404 18.0 0.0202 0.50
17 13.66 25.0 43.0 .0170 .43
.0454 14.5 .0212 .47
18 14.74 25.0 39.5 .0132 .33
.0012 37.5 .0012 (1.00)
19 14.77 12.5 50.0 .0130 .16
.0232 23.0 .0138 .59
20 15.32 12.5 35.5 .0102 .13
.0016 33.5 .0016 (1.00)
21 15.36 25.0 58.5 .0098 .24
.0092 38.0 .0072 .78
22 15.58 25.0 63.0 .0090 .22
.0004 61.5 .0004 (1.00)
23 15.59 12.5 74.0 .0090 .11
.0046 55.0 .0040 .87
24 15.70 25.0 80.0 .0102 .26
.0420 18.5 .0160 .38
25 16.70 12.5 31.0 .0108 .14
.0012 30.0 .0008 (.67)
26 16.73 25.0 55.0 .0106 .26
.0072 39.5 .0060 .83
27 16.90 12.5 52.0 .0104 .13
.0012 49.0 .0012 (1.00)
28 16.93 25.0 74.0 .0104 .26
.0310 22.5 .0146 .47
29 17.67 25.0 47.5 .0106 .27
.0012 45.0 .0012 (1.00)
30 17.70 12.5 57.5 .0104 .13
.0088 38.0 .0070 .80
31 17.91 25.0 63.0 .0104 .26
.0008 60.0 .0008 (1.00)
32 17.93 12.5 72.5 .0106 .13
.0302 22.5 .0148 .49
33 18.65 12.5 35.0 .0108 .13
.0004 34.5 .0004 (1.00)
34 18.66 25.0 59.5 .0106 .27
.0106 36.5 .0082 .77
35 18.91 12.5 49.0 .0106 .13
.0004 48.0 .0004 (1.00)
Exploring the Variety of Random
Documents with Different Content
“Come in,” called a voice.
Hollister opened the door and stood hesitating on the threshold.
Hildebrand sat alone by the table, and, as he glanced up and saw
who his caller was, his face darkened.
“Well?” he said curtly.
Hollister flushed and took a step forward.
“I—I’ve been—a fool, Blair,” he stammered. “I beg your pardon for
what I said this afternoon.”
“Oh, you’ve found that out, have you?” Hildebrand inquired
sarcastically.
He was still sore over the result of his attempt that afternoon to
open Hollister’s eyes as to the real character of Blake. It had not
been a pleasant nor an easy thing to do, and Bob’s reception of it
had cut him to the quick, besides making him furiously angry.
“Yes; he’s all you said of him and more,” Hollister returned in a low
tone. “I just found out, and I couldn’t rest until I had told you how
sorry I am about the way I talked to you.”
His manner was so dejected, and the look of penitence in his eyes
was so very real as he turned toward the door again, that
Hildebrand could not help but relent.
“Come back here, you old idiot!” he exclaimed, springing to his
feet. “You certainly did made me hot this afternoon, but what’s the
use of keeping mad? Give us your fist, and the next time don’t be so
infernally set in your way.”
Hollister’s eyes brightened as he gripped the proffered hand.
“You’re all to the good, Blair,” he said quickly. “Most fellows would
have felt like kicking me downstairs.”
“I felt worse than that this afternoon,” the big guard grinned. “But
nobody can stay mad with you very long, Bobby. Sit down and let’s
hear about it.”
Hollister told the story briefly, and then, in spite of his friend’s
urging, he departed to put in the rest of the evening in hard
studying. Since it was the first time he had really applied himself to
his books in weeks, he naturally did not make much progress, but at
least it was a beginning.
The blow came the next morning, when the first mail brought him
a letter from the dean’s office. He opened it with trembling fingers
and glanced through the brief contents. The typewritten
communication was short, terse, very much to the point, and bore
the scrawly signature of the dean himself.
“Dear Sir: Since you have seen fit utterly to disregard my advice of a
week ago, I am forced to tell you that unless you attain a grade of at
least sixty in every recitation from now until the beginning of the winter
vacation your name will be dropped from the rolls of the senior class.”
In perfect silence, jaws set and face a little pale, Hollister read the
short note through the second time.
“Holy cats!” he muttered. “That’s the end of yours truly, all right!
Sixty per cent.! Why don’t he say a hundred and be done with it? I
stand about as much show of getting it.”
Now that it was too late, he saw with vivid clearness the extent of
his amazing folly. Merriwell had done his utmost to make him realize
the seriousness of his position a week ago. Jim had been trying his
best to help him for a longer time than that. Even the dean had
strained a point of college discipline in his favor. And in spite of all
this he had gone his way blithely and blindly, living only in the
present, with a perfectly suicidal disregard for the future.
What could he do? What was there possible for him to do? He was
in despair. He had no more than a glimmering of the work for that
day. It would need nothing less than a miracle for him to get the
required percentage.
The more he thought over the matter, the more despondent he
became. At length, as a last resort, he resolved to go to Dick with
his troubles. He did not hope for any happy solution of the difficulty,
but there is always a little comfort in talking over one’s miseries with
somebody; and Bob knew that Dick would never say, “I told you so.”
Happily, the first recitation was scheduled for eleven o’clock, and
Hollister found Dick alone in his rooms working over some math
problems. He looked up smiling as the dismayed fellow entered.
“Hello, Bobby,” he greeted. “What’s the matter? You look as if life
held no further joys for you.”
Without a word, Hollister thrust the dean’s letter into Merriwell’s
hand. Dick read it through with knitted brows, and, having finished,
folded it methodically and handed it back.
“Wouldn’t that kill you dead!” he exclaimed. “Sixty per cent.! Let’s
see how we can dope that out.”
Hollister looked at him blankly.
“Dope it out!” he exclaimed. “What is there to dope out? I’m
done!”
“Rot!” Dick returned emphatically. “You’re not going to give up
without an effort, are you? We’ll get you through somehow. But
you’ll have to buckle down and work like a terror.”
“I’ll work, all right,” Hollister returned, in a dispirited voice; “but I
can’t make that average. Why, I’ve got to start in and make it this
very day, man, and I haven’t the haziest notion of what the Latin
lesson is, though I did grind some on chemistry last night.”
“Never know what you can do till you try, Bobby,” Dick said
cheerily. “Why, we can’t let you be dropped, old fellow. Rather than
that, I’ll turn tutor and drag you through by the hair of your head.”
He paused and his face grew serious.
“There’s one thing sure, though,” he went on, his eyes fixed on
Hollister’s face; “you’ll have to give up football, and drop it like a hot
cake this very day.”
For an instant Hollister looked at him blankly as if he did not
comprehend what the other had said. Then he understood, and a
look of utter despair came into his eyes.
“Give it up!” he cried. “Oh, Dick, I can’t!”
“You’ve got to,” Merriwell retorted firmly. “Can’t you see that if you
don’t you’ll be dropped sure as fate? You can’t play football and
study at the same time. You’re not made that way. It’s a question of
giving it up voluntarily or of being dropped from the class and,
consequently, from the varsity.”
Hollister groaned. How could he give up the thing he loved better
than anything else in the world! What would college life be without
it? He almost felt as if he’d rather be dropped than voluntarily give it
up, except that such a course would mean the same thing in the
end.
He looked at Merriwell pleadingly.
“But I could still play in the games, even if I didn’t show up for
practice, couldn’t I?” he urged.
Dick shook his head.
“You couldn’t,” he said decidedly. “You’ve got to the point when
you have to give every atom of your mind to your work. The minute
you begin to think about playing in a game your attention will be
distracted. You won’t be able to study. It can’t be done, Bob. You
don’t suppose I’m anxious to see you leave the team, do you? Great
Scott, man! I don’t know what we’ll do without you. But it’s your
only chance. Don’t you see that?”
Hollister saw it only too clearly. He realized perfectly the truth of
Merriwell’s words. He knew quite well that if he were going to play in
a game he would be thinking for days beforehand about it.
Unconsciously his mind would wander and he would cease giving the
proper attention to his books. Bitterly he regretted the moment
when he first began to let things slide. If he had only not let his
enthusiasm for the game get the better of him he would be all right
now.
And suddenly into his mind came the thought of Jarvis Blake and
his treachery. The fellow would triumph now and would very likely
get his place on the varsity. He could not bear the idea.
“If I quit the team Blake will be put on,” he said aloud. “I couldn’t
stand that, Dick. It’s what he’s been after right along. Last night—I
heard——”
A gleam of combat came into Merriwell’s eyes and his chin
squared.
“I thought so,” he said emphatically. “I had a notion that was his
game. But it won’t work if I can put a spoke in his wheel. There are
a couple of other subs who are as good as he is. I rather think one
of them will take your place.”
“If you could only work it, Dick!” Hollister said eagerly. “Of course,
I’m not trying to blame him for what’s happened. That’s all up to
me. But I do know that he did his best to have me dropped, and if
he got my place in the line I couldn’t stand it.”
“Don’t worry,” Merriwell said quickly. “I don’t think he will.”
He paused and looked Hollister keenly in the eyes.
“Well,” he said slowly, “have you made up your mind?”
Still Bob wavered, unwilling to take the step which, deep down in
his heart, he knew would have to come.
Merriwell showed no signs of impatience. With rare sympathy, he
realized what a struggle must be going on in the man’s mind. The
thought of all it would mean to him if, for any reason, he were
forced to give up football was appalling, and he knew that Hollister
was even more devoted to the game.
“I know how hard it is, Bobby,” he said quietly. “But after a little
you’ll come to see that it’s the only thing for you to do. Football—any
game, in fact—is a splendid thing when it keeps its proper
proportions as something incidental to the college course. But the
minute it begins to dominate a man, as it has done you to the
exclusion of everything else, it’s time to cut it out. You didn’t come
to Yale to play football, but to get your degree and the other
benefits which a college course gives a man. Think how you’d feel if
you were dropped at the very beginning of your senior year. Think of
the humiliation of being thrown out with such a record as you have
made this fall.”
“I can’t even play in the Yale-Princeton game on Saturday?”
Hollister questioned sadly.
Dick shook his head firmly.
“No, sir,” he returned with emphasis. “You give me your promise
never to play football again while you’re in college, and I’ll do my
very best to pull you through in your studies. How about it?”
“All right,” Hollister said, in a low voice. “I promise.”
“Good,” Dick smiled. “That’s the stuff. Now let’s get down to
business.”
He glanced swiftly at the clock.
“An hour and a half before Latin,” he murmured. “We’ve got to get
busy.”
Before Hollister knew what he was doing, Dick had him sitting at
the table, the open book before him, and together they proceeded
to go through the day’s allotment of Horace.
Merriwell did his work thoroughly, translating slowly and stopping
to explain the derivation of every word about which Bob had the
least doubt. He had a natural gift of making things plain, and in an
hour’s time Hollister had acquired a pretty good notion of what it
was all about. Then, after a hurried review of the chemistry lesson,
they sallied forth to the lecture room.
“I think you’ll do in the Horace, old fellow,” Dick assured him. “Just
keep your head and take it slowly, and you’ll come out all right.”
Such proved to be the case. About halfway through the hour,
Professor Goodhue called Hollister’s name in a rather weary tone of
voice, fully expecting a repetition of the absolute failures for which
the fellow had become noted.
To his amazement, Hollister arose slowly and gave a very good
rendering of the passage, even to construing accurately the few
words the dazed professor asked him.
“That will do, Mr. Hollister,” the latter managed to say when Bob
had finished. “Very good indeed. I should—er—like to congratulate
you on the extraordinary improvement in your work.”
“Thank you, sir,” Bob murmured, his face a bit red.
On the campus outside, Dick slapped him on the back.
“Well done!” he exclaimed. “That was more than sixty, all right.
You’ll do. Now for the lab. That’s going to be harder, for we didn’t
give any time to it.”
As they mounted the steps to the chemical laboratory, Bob
happened to catch a glimpse of Blake’s face, and the look of ill-
tempered annoyance he saw there was an added incentive to
renewed endeavor. The big, blond fellow was evidently not at all
pleased with the surprising turn things had taken.
By some fortunate chance, Hollister was not called upon at all in
chemistry. Perhaps the professor had grown weary of his constant
failures and did not think it worth while. At all events, it gave Bob a
little respite. There were no other recitations that day, and by to-
morrow, he hoped, with Dick’s assistance, to have made up a little of
the lost time.
Merriwell realized perfectly that what he had undertaken was not
going to be any easy task. There was no fun at all in coaching a
fellow who had done absolutely no work for almost six weeks, and
was, consequently, totally ignorant of what had been gone over so
far that term. But this fact did not deter him in the least. He knew
that it was the only way by which Hollister could be saved, and,
though it meant that every spare moment must be devoted to
tutoring Bob for a few weeks at least, he was fond enough of the
fellow to go to that extreme.
Hollister’s announcement at the training table that he had to leave
the team was one of the hardest things he had ever done. It had the
effect of a bombshell on the assembled players.
Instantly the room was in an uproar. The fellows all crowded
around him, unable to believe their ears.
“You can’t leave, Bobby!”
“Cut it out, old man, and have another think.”
“Stop your kidding!”
“Thunderation! What’ll we do without you?”
These and a dozen other incredulous exclamations were hurled at
the wretched fellow, but Bob persisted in his resolve; and when the
men saw that he was really in earnest, they were in despair.
All, that is, save Jarvis Blake. Dick, his eye on the fellow, noticed
the sudden expression of amazed incredulity which flashed into his
face, to be followed instantly by a look of joy and unmistakable
triumph. Evidently he had not expected this turn of affairs, but he
was none the less more than satisfied with it.
“I’ll put a spoke in your wheel, my bucko,” Dick muttered fiercely.
“All your dirty scheming won’t do you a bit of good.”
He put in an hour’s work with Hollister after dinner, and, laying out
enough to keep the man busy that afternoon, he got out the car and
drove down to the field.
His first move was to seek out Tempest and Bill Fullerton, and for
ten minutes the three men remained in close confab. When they
separated there was a look of extreme satisfaction on Dick’s face. He
hurried into the athletic house to get into his togs.
A little later, when the men were all assembled on the field, Don
Tempest held up his hand for silence.
“You fellows all know that Hollister has been obliged to leave the
team,” he said quietly. “You also know why. It’s something which
can’t be helped, but I’m sure you will agree with me that it hits us
pretty hard and will make a big hole in the line. I’m sorry it couldn’t
have been postponed until after the game on Saturday, but since
that was impossible we’ll just have to make the best of it. In regard
to filling his place——”
He hesitated and his eyes wandered over the eager, expectant
faces of the subs. Many of them knew that there was no possible
chance of their being picked for the important position, but there
were three or four who evidently had hopes.
Jarvis Blake had more than hopes, if one could judge from the
look of assurance on his face. There was plainly small doubt in his
mind that he would be the lucky man, and Dick watched him with a
distinct feeling of satisfaction.
“In talking it over,” Tempest continued, “we have decided that
Keran had better try out for end until further notice.”
Blake gave a gasp of dismay. The blow was so sudden and so
absolutely unexpected that, for an instant, he could not believe his
ears.
Then his face turned scarlet, his eyes flashed, and he took a quick
step forward. Dick was watching him quietly.
“I think——” began the big, blond fellow, speaking with evident
difficulty.
Tempest eyed him coldly.
“I said Keran,” he remarked significantly; “Phil Keran.”
There was an undercurrent of contempt in his voice which cut
Blake like the lash of a whip and made him step back involuntarily.
Before he could recover his customary poise, the fellows spread out
in the regular formation, Keran, grinning from ear to ear, in the
coveted place at right end.
Blake had never been so furious in his life. He could not
understand how it had all come about. For a moment he was
tempted to leave the field. He had even turned and was about to
stride off without a word, when he realized that such a move would
be folly. He would gain nothing by it, and his chances for ever
accomplishing his end would be totally ruined.
With a sullen scowl on his face, he walked over to his place on the
scrub. After all, Keran was only in the varsity on sufferance. He
might not make good, and then Blake’s chance would come.
CHAPTER VII
THE BEGINNING OF THE GAME.
It must not be supposed that Bob Hollister’s course was an easy

one. It was, on the contrary, desperately hard. A dozen times a day
bitter thoughts and regrets for what he had given up assailed him,
but he managed to thrust these aside, and, with Dick’s help, he kept
doggedly at his work, encouraged by the very evident progress he
made in his studies.
The story of his renunciation of football and his steady application
to his books seemed to have become known to the faculty. Certain it
was that, one and all, they realized what an effort he was making to
stick with the class, and most of them did their best to help him
along.
As for Merriwell, every minute he could spare was devoted to
coaching Bob. The latter almost lived in Dick’s rooms. Every evening
they went over the work for the next day together, Dick patiently
explaining every point, bolstering up Hollister’s failing courage,
making a regular hermit of himself for the sake of the other man’s
future.
In the afternoons Bob spent his time grinding on the back work,
for occasionally the professors had an annoying way of having little
quizzes which covered the subjects they had gone over that term.
That was the hardest part of it all, to sit alone with a book before
him, knowing all the time that the others were out on the field
where he longed to be more than anywhere else in the world. At
first he had to grit his teeth and exercise the utmost self-control to
keep his mind from wandering; but, after a little, it came easier,
though he was never wholly resigned.
At last came the day of the Yale-Princeton game. Hollister
wondered desperately whether he would have to stay away from the
field that afternoon. It seemed as if that would be more than he
could bear. In the morning he broached the subject to Merriwell.
“About the game this afternoon, Dick,” he began hesitatingly. “It
don’t seem as if I could study while that’s going on. Couldn’t I go
down and watch it, just this once?”
Dick looked at him thoughtfully.
“Do you think that’s a good idea, Bobby?” he asked slowly.
“Wouldn’t you feel worse on the field, not being able to play, than
you would if you stayed away?”
“Gee, no!” exclaimed Hollister. “Even if I don’t play, there’d be
some satisfaction watching it.”
“Come on, then,” Merriwell said quickly. “You’ve certainly done
well enough to take the afternoon off.”
Thus it was that Hollister sat in the tonneau of Dick’s car as the
Wizard tore down to the field that afternoon. Tempest and Blair
Hildebrand sat with him, Rudolph Rose crumpled his long legs in the
body of the car at their feet, while Teddy Baxter clung precariously
to the running board.
Hollister felt a thrill of the old joyful enthusiasm as the car whirled
through the streets. Once more he seemed to be one of them, and,
as he entered the grounds and swept his eye over the already filling
stands, he sniffed the air like a war horse that scents combat from
afar.
But once in the dressing room, the reaction came. He saw the
others strip and hurriedly don their togs; listened to their eager,
excited discussion of their chances for victory; watched them troop
out in a body and lope across to the gridiron; and, as he followed
slowly, dispiritedly, he realized with a bitter pang that he was out of
it. Instead of plunging into the contest with tingling blood and every
sense alert, doing his best for his Alma Mater, straining every nerve
to win a victory for the blue, he must stand on the side lines and just
watch.
The thrilling, deep-toned cheers of the excited thousands would
ring in his ears as before, but they would have a different sound.
They would be meant for others, not for him. Somehow, he felt that
if he could only have played in this one game he could be resigned
about never going on the field again. If he could only show just once
more what he could do—play just one more game for all that was in
him, and perhaps help to win a victory, it would content him.
But it was too late. He had given his word, and the team was
finally made up. With downcast eyes and bitter heart, he entered the
inclosure and, walking past the grand stands, dropped down on the
side lines with the subs. At least he would watch the game from the
field. He couldn’t bear sitting in a stand. He had never done that in
all the time he had first come out for the team.
The stands were filled to overflowing, a sea of eager, enthusiastic
faces rising, tier upon tier, from the field. Flags fluttered by the
hundreds, blue, mostly, but with a liberal sprinkling of the orange
and black. The hum of many voices sounded like the drone of a
gigantic hive of bees. The flash of many faces turned impatiently
toward the closed gates as the hour approached.
At last the gates were flung open and the teams appeared.
Princeton came first, and cantered briskly across the field. They
were greeted by a round of applause from their adherents.
Then Yale appeared, and the stands rose to them with a yell which
sent a thrill through Hollister’s heart—a thrill followed swiftly by a
stab of pain. Perhaps Dick had been right when he said it would be
harder here than if he had stayed away.
Yale won the toss, and, there being a rather brisk wind blowing,
chose the protected goal and gave the enemy the ball. The fellows
swiftly took their places to await the kick off. Presently the whistle
sounded, and from that moment Bob Hollister was oblivious to time
and space, the shouting crowd, the excited subs—everything, in
short, except the progress of the contest before him.
Almost at once he saw that Princeton had an unusually strong
team. He had expected something of the sort, for all reports agreed
in stating that it was the best eleven the New Jersey college had
turned out in several years; but Hollister had not thought it would be
quite so good as it now appeared.
With knitted brows, he watched the progress of the ball down the
field toward Yale’s goal. There was no doubt in his mind that the
orange-and-black fellows had made the most of some very efficient
coaching. Their teamwork was splendid, and every now and then
they made use of some novel play which caused Hollister to bestow
upon them a sincere, if somewhat grudging, admiration.
But presently he ceased to watch their good points and bent an
anxious, scrutinizing eye upon his former comrades. Something
seemed to be the matter with their playing. A subtle, impalpable
something, hard to define, but plainly evident to the quick mind of
the man on the side line.
There was a slight absence of snap, of unity, which perhaps
another might not have seen. Hollister was entirely too modest to
realize that his absence from the team could make any difference.
He did not see that the lack of his swift, perfect brainwork, his
cheering encouragement, would be felt to any appreciable extent.
And yet, that was actually the case.
Merriwell was playing a perfect game, Buckhart was at his best;
but they could not carry the whole team. Don Tempest, still not
perfectly strong after his long illness, and feeling the lack of the
practice which he had lost, did not make a very good showing. While
Phil Keran, though he was a good steady player and did his best,
could never take the place of Hollister, one of the best ends Yale had
ever had.
Slowly the ball was forced back. Nearer and nearer it came to the
goal. Bob’s heart leaped into his throat and he could not swallow.
They must not make a goal—they must not!
Then the line stiffened, the advance ceased. Two downs brought
barely five yards gain. Not daring to risk another forward pass,
Princeton tried a kick from the field.
The ball soared over the heads of the scrimage line. To Hollister,
tense, breathless, it seemed as if it would pass over the bar, and he
groaned aloud as the orange-and-black line surged forward in its
wake.
The groan changed to a gasp of joy as the pigskin carromed from
an upright and a tall, lithe figure leaped into the air, clutched it and
dropped back.
It was Merriwell. Bob could have shouted aloud in his relief had he
not been too intent on watching the outcome. For an instant the
men were so involved in a tangle of flying figures and waving arms
that he could not see what had become of the ball.
Then, all at once, a man darted around the end, closely followed
by two others, and sped over the ground in an oblique course
toward the farther side line.
In an instant Bob recognized him as Crowfoot, and realized that
Dick had in some way passed the ball swiftly to the Indian, who,
assisted by Elwell and Kenny, the quarter back, was covering the
ground like a streak of light.
Kenny was bowled over instantly; Elwell met his Waterloo a
minute afterward; but by the time Crowfoot was tackled by one of
the Princeton guards he had covered thirty yards and the ball was
back out of danger.
Then the whistle sounded and Hollister realized that the first
quarter was over.
After the brief three-minute interval, Yale started in with a rush,
carrying the ball down the field in a series of brilliant plays which did
full credit to every man on the team.
They seemed to have recovered from their strange lassitude and
were evidently determined to utterly annihilate their opponents.
But that was not to be done easily. Oddly enough, Princeton
blandly refused to be annihilated. And so the hard-fought battle
continued. Back and forth surged the lines of tattered, gasping,
breathless men. At one moment it would seem that Yale had the
advantage, and apparently nothing could prevent her from scoring.
Then Princeton would rally and force the blue line slowly, but surely,
back from the danger zone.
To the man on the side line it was sheer agony. His trained eye
saw the weak points of his team even more swiftly than did
Tempest, the captain. His alert brain, feverishly active, took in lost
opportunities which the men on the field did not even perceive, and
he was constantly thinking of how he would have made a successful
play if he had only been out there with the rest.
Then began a series of minor accidents which played havoc with
the Yale line. First of all, Rose was knocked senseless and had to
leave the field. Then Samp Elwell twisted his ankle so that he could
not stand on it; and another sub threw off his enveloping blanket,
jerked off his sweater, and raced into the arena in response to
Tempest’s peremptory gesture.
Last of all, Phil Keran gave out, and, after a momentary hesitation,
Tempest reluctantly summoned Jarvis Blake from the side line. He
was the best man left, and, perhaps, had it not been for what he
had heard from Dick about the fellow, Tempest might have put him
in before; for Blake had always showed up well in practice.
As Hollister saw his enemy race out and take his own place at
right end, he clenched his fists so tightly that the nails cut into the
flesh of his palm. This was the worst of all. Blake was now just
where he had been scheming to get.
Then the teams lined up and Bob forgot even that. It became
apparent at once that the change had not been for the better.
Princeton had been obliged to put in only one substitute, and her
advantage showed very plainly.
Strive as the Yale line did against them, the solid phalanx of the
opposing team made its way inexorably down the field. There were
occasional rallies, to be sure, but never once did the orange and
black fail to make their required gain; and at last, with a sob in his
throat, Hollister saw the pigskin forced over the line and heard the
Princeton crowd thundering its joy.
The goal was kicked, and, before the second quarter was over,
Princeton had scored again on a drop kick, and was nine points to
the good.
Things looked very black for Yale.
Hollister did not leave his place on the grass. He could not bring
himself to go back to the house with the team. He had not the
heart. And so he lay there viciously jabbing the blade of his knife
into the ground, his brow drawn into a scowl, his brown eyes full of
a strange mixture of longing and pain.
He had been watching Blake’s playing, and it had taken him only a
few moments to see how much it fell short of his own. Hollister was
not in the least conceited, but he had a keen sense of sizing a fellow
up on the field and had always viewed his own good points and
shortcomings as dispassionately as he did those of any one else.
Watching Jarvis Blake, he knew that he himself could have done
better. Blake was a good player, but he was deficient in some
important qualifications, principally initiative and speed in starting.
Time and time again, Bob saw him fail to take advantage of an
opportunity which might have meant a gain of yards to his team.
Once, in his excitement, he had shouted a warning to the substitute,
only to realize what he was doing and choke himself into silence.
The third quarter started off with a fresh swing. The rest had done
all the men good, and evidently there had been some straight talk in
the athletic house which heartened them and brought them to a
realizing sense of the gravity of their position.
The ball was forced down to within the thirty-yard line without a
pause. Hollister, watching eagerly, soon saw whose brain was
dominating the work. Almost every time the pigskin was passed to
Merriwell. And, with quite as much regularity, the brilliant senior
responded nobly.
He seemed to be everywhere at once, slippery as an eel, dodging
hither and thither in a most bewildering fashion, sometimes passing
the ball to Crowfoot, or another on whom he could depend, but
always making gains, ever advancing, until Bob found himself sitting
erect, his cheeks burning and his eyes sparkling as he watched this
amazing exhibition of almost perfect football.
Would he make it? Could he possibly hold out to reach the line?
Suddenly his question was answered.
The quarter back ripped out a rapid signal which Bob could not
hear perfectly; the ball was snapped back; there was a bewildering,
lightninglike, intricate pass. Hollister gasped. It was his improved
crisscross play, the last thing he had worked out before he had left
the team.
The pigskin seemed to leap from one man to another like a thing
endowed with life. For a minute he lost track of it, and then he
caught his breath swiftly as Merriwell sprang out of the mêlée, the
pigskin tucked under his arm, and raced over the turf as if he were
as fresh as the moment he had first set foot on the field.
The Princeton crowd was taken by surprise. The pass had been so
cleverly made that most of them thought the ball was being sent
around the other end, and there was a surging rush in that direction,
which left a comparatively free field for Dick.
Too late they saw their error and trailed after him.
There were but two men between him and the coveted goal. He
could easily outdistance the first, who was a little to one side, but
the full back would have to be dodged.
As he ran, he watched the man keenly, wondering just what trick
he would have to bring into play to get away from him. The fellow
stood alertly on his toes, watching, waiting, ready to spring to one
side or the other, as the case might be.
Dick came on without slackening his speed, swerved suddenly to
the right, whirled, darted the other way, and all in such a brief
moment that to this day Princeton’s full back hasn’t the least notion
of how he was fooled. He only knew that by the time he had turned
Dick was a dozen feet away, speeding on toward the goal.
The next instant the full back gave a grunt of triumph and
stretched himself, for the Yale man suddenly staggered, tried wildly
to recover, and then fell full length to the sod.
A groan of horror went up from the stands, followed by deathlike
stillness.
Then, to the amazement of the onlookers, they saw that, instead
of lying where he had fallen, Merriwell spun end over end, and the
next instant he was on his feet again. But he ran with an appreciable
limp.
It was a tense moment. The full back was gaining. Slowly, but
surely, he crept up and the distance between the two lessened. Dick
ran with more and more apparent effort, and it was plain to all that
he must be suffering tortures.
Now the full back’s fingers touched him, but could find no hold on
the smooth canvas. The next instant they clutched his waist, and
clung there with a firm, dragging grip.
Five yards more! Could he ever make it?
Struggling, dragging, straining every nerve and muscle, Merriwell
flung himself over the line; and, as he did so, a great sigh arose
from the spectators, merging into a crashing burst of sound, for they
realized that the ball was over.
CHAPTER VIII
A BROKEN PROMISE AND A VICTORY.
Despite his sprained ankle, Merriwell kicked the goal, straight and
true, and the teams lined up again. But that run had been a last
desperate attempt to wrest victory from defeat.
Unable to count longer on Dick, who, though he was still able to
play, could not be expected to continue the extraordinary efforts
which had made him an object of wonder to every man on the field,
the team went to pieces as nearly as any Yale team can.
They played despairingly, doggedly, disputing every inch on the
part of the Princeton organization, but for all that being borne slowly
down the field.
The ginger was gone out of them. They had no life, and their
playing had become more or less machinelike.
Bob Hollister realized this swiftly. He knew the signs only too well.
“They can’t do it!” he almost sobbed. “They can’t beat them that
way!”
If he could only go into the game. Just for that last quarter. Surely
it could not do any harm. He must do it. He could not sit there and
see the fellows beaten.
The third quarter was nearly over when he leaped to his feet, his
face white and determined, and ran swiftly toward the house.
Dashing inside, he encountered Keran, his face a network of
scowling lines, his fists clenched, and one foot tied up in bandages.
“Gimme your clothes!” Hollister exclaimed. “Quick!”
“What——” gasped Keran.
“Blazes!” ripped out the excited fellow. “Your clothes, I tell you!
Get ’em off! Mine aren’t here!”
With an exclamation of joy, the other realized what he meant to
do. Snatching off his jacket and jersey, he tossed them to Bob, who
was already half undressed.
“Glory be!” he cried. “You’re going to play! You’ll brace ’em up!”
Hollister made no answer. His eyes were gleaming. One thought
only was in his mind. He must get into those togs and back to the
field before the beginning of the last quarter. He meant to play if he
never did another thing in all his life. His promise to Merriwell was
forgotten. He thought of nothing but that line of gasping, tattered
men out there, striving vainly against black defeat.
With eager, trembling fingers, Keran helped him lace his jacket.
Rudolph Rose staggered up from where he lay full length on a
bench, and, dropping down on the floor, laced up his shoes. Neither
of them spoke a word, for words were unnecessary. They
understood.
In a miraculously short time Bob was ready, and, snatching up a
nose guard, he tore out of the house.
Bill Fullerton, his face black as a thundercloud, was talking to
Tempest on the side lines. The brief intermission was almost over as
Bob dashed up to them.
“I want to go in, Don!” he exclaimed.
Both men looked at him in astonishment.
“I thought——” Tempest began.
“Never mind that,” Hollister interrupted. “I’ve got to go in! That’s
the only way. The fellows have gone all to pieces since Merriwell hurt
himself!”
Still the captain of the varsity hesitated. He knew quite well of the
promise Hollister had made Merriwell that he would not play football
again during his college course.
“I swear to you, Don, by all that’s holy,” Bob said earnestly, “that if
you let me play out this game I’ll never touch football again! It’s only
fifteen minutes, Don! Just fifteen little minutes! If I sit here watching
it, I shall go mad. Let me play, Don.”
His pleading voice quivered with the emotion which was tearing
him.
Tempest was in somewhat of a quandary. He wanted to put
Hollister in, for he felt that it was barely possible that Bob might
succeed in putting spirit into the jaded, discouraged men. He was
fresh, too, and wrought up to a white heat of enthusiasm. It would
be strange if he did not accomplish something. Don glanced at
Fullerton questioningly.
The coach nodded emphatically.
“It’s the only thing that can possible save the day,” he said
decidedly. “Better let him in.”
“Who——”
“Blake, of course!” Fullerton said tersely. “He’s rotten!”
Hollister’s face lit up joyfully as he listened to this brief
conversation. Then the signal came, and there was a general
movement to get out on the field.
Tempest walked rapidly to Blake’s side and said a few words to
him in a low tone. The big, blond fellow flushed scarlet and darted a
venomous glance at Bob. Then, without a word, he turned on his
heel and walked rapidly toward the athletic house, his face sullen,
and the angry flush still in his cheeks.
Hollister followed the other men with a springy step and a heart
fairly bursting with joy. At last he was back with the boys. It seemed
almost as if he had never left them. He did not worry over the fact
that, after these brief, fleeting minutes were over, he could never
play again. He only knew that the team was in a bad way and
needed him, and he resolved that he would play as he had never
played before.
One after the other the fellows recognized him and greeted him
with short, hurried words, which were an odd blending of surprise,
joy, and relief; but all had such a ring of sincerity and truth that
Hollister was more touched than he would have thought possible.
He dared not meet Merriwell’s glance. He had broken his promise,
and he was not sorry; he hated to think of what Dick’s opinion of
him would be from this time forth.
Then, as he crouched in his place, he forgot Merriwell, forgot
everything but the fact that he was back in the line again.
“Are you all ready?” asked the referee.
There was no reply. Only here and there a foot moved uneasily as
weights were thrown forward, and there was a general, almost
imperceptible, tightening of nerves and muscles.
Then the whistle shrilled.
Those who watched the game that day said afterward that, in all
their experience, they had never seen such an amazing rallying on
the part of any team as was shown by the Yale eleven during that
last quarter.
Three minutes before they had gone off the field with dragging
steps and gloomy, discouraged faces. The followers of the blue, who
crowded the stands, felt a wave of despair sweep over them as they
thought of what might happen in that last fifteen minutes. Many of
them fully expected to see Princeton make another touchdown, if
not two, and they waited with perfunctory, mechanical cheers, and
swiftly ebbing spirit for the beginning of the end.
But the sudden, totally unexpected appearance of Hollister
seemed to work almost a miracle.
Bob responded nobly. Never had he put up such a game before.
Tireless, never failing, swift as lightning, with his brain in splendid
working order, he seemed to be all over the field at once. Dodging,
slipping through holes in the line where one would not have thought
any advance possible, blocking, cutting off opposing runners, and
interfering for runners of his own team, it seemed as if all the pent-
up, thwarted energy of the last few days of deprivation was being
poured out now in this brief, brilliant exhibition.
His work thrilled the other men with a new hope, and stirred them
to fresh endeavor, so that they were with him heart and soul; and
the pigskin was rushed down the field swiftly and irresistibly, until
the forty-yard line was reached.
Here the orange-and-black fellows seemed to recover, and,
rallying, presented such a solid line that two downs brought barely
six yards; and Yale had to resort to a drop kick, which sent the ball
forward thirty yards, but gave it to Princeton.
Then the great struggle of the day began. Inspired by the brilliant
Hollister, Yale made a strenuous, dogged effort to score, while her
opponents were equally determined that she should not. Back and
forth surged the lines of men, never reaching within kicking distance
of either goal, and using up the precious minutes in fiercely
contesting every inch of progress.
It was a battle royal, and the spectators were so thrilled with
interest and excitement that they almost forgot to cheer.
At last, when there were but six minutes left to play, Kenny
decided to make use of one of the most intricate and most daring of
the combinations of double plays and crisscrossing which the
coaches had worked out from Hollister’s suggestion. It was only to
be used as a last resort, and Kenny decided that the time had come.
“Sixty-seven—twenty-four—thirty-two——”
Kenny paused. Merriwell sprang back a yard. Buckhart crept a few
feet in.
“Fifty-four—seventeen!” finished Kenny swiftly.
The ball was snapped, Brad ran forward three strides, Kenny
turned, and the pigskin flew back. The next instant Merriwell had the
ball, and sped toward the right end of the line. The quarter crossed
in front of him; the tackle and guard thrust back their opponents;
the Princeton line surged forward with a rush.
Hollister plunged forward, too, as if he were intent only on
interfering in Merriwell’s behalf; but he had a more important duty
than that to perform. Swiftly, before their opponents realized what
was being done, he and Dick changed places, Merriwell was blocking
with all his might, while Hollister, the ball clutched tightly to him,
sped round, shot through and out onto the field, leaving a mass of
waving legs and arms many yards behind.
Joy was the supreme sensation in Bob’s breast. Only the Princeton
full back threatened. The ball was safely clutched in his right arm,
his breath came easily, his legs were strong, and the goal posts
loomed down the field and beckoned him on. This, he thought
exultingly, was the best moment that life could give.
Behind, although he could not hear it for the din of shouting from
the stands, he knew the pursuit to be in full cry. He edged farther
out from the dangerous touch line and sped on. The Princeton full
back had been deceived by the play, and had gone farther up the
field for a kick, and now down he came at full speed.
Hollister seemed to hesitate and falter. The full back prepared to
tackle. His broad back was bent far over, his sturdy legs squared
themselves, and, when Bob was almost within his reach, he dove
forward.
There was a sudden gasp from the spectators, a breathless hush,
and then a thunderous roar of joy, as Hollister leaped high in the air,
cleared the hooking arms, stumbled, got his balance again, and ran
on, free, the ball still cupped in the curve of his arm.
The momentary pause had served to bring the foremost of the
other pursuers almost to Bob’s heels.
And now the plucky end began to feel the effects of his strenuous
work. His breath came irregularly, his throat was parching, his legs
ached with every bound, but still he never wavered. Behind him
sounded the thud of relentless feet. He dared not look back lest he
stumble. Every second he expected to feel the clutch of the enemy.
Presently he gave up trying to breathe; it was too hard. His head
was swimming and his lungs seemed bursting.
Then his wandering faculties rushed back at a bound as he fancied
he felt a touch—just the lightest fingering—and, gathering all his
remaining strength, he increased his pace for a few steps.
The ten-yard line passed, slowly, reluctantly.
“One more,” he thought. “Only one more!”
The great stands were hoarse with shouting, for here ended the
game.
Nearer and nearer crept the five-yard line; nearer and nearer crept
the pursuers. Once more Hollister called upon his strength, and tried
to draw away, but it was useless. And, with the goal line but four
yards distant, stout arms were clasped tightly around his waist.
One—two—three strides he made. The goal line writhed before his
dizzy sight. Relentlessly the clutching grasp fastened tighter and
tighter about him like bands of steel, and settled lower and lower
until his legs were clasped and he could move no farther.
Despairingly he thrust the ball out at arm’s length, and tried to throw
himself forward; the trampled turf rose to meet him, and then
blackness came.
Bob’s first waking thought was that he must be back on the rocky
shores of Maine, where he had spent the past summer. Surely those
were breakers which roared and thundered in his ears. Then he
opened his eyes, and found that he was lying on the sod, a sweater
under his head, and several vaguely familiar faces swimming above
him.
A moment later he knew that it was not surf, but the wild yelling
and cheering of excited, enthusiastic thousands. Back and forth
rolled the mighty torrents of sound, breaking and crashing in
reverberations.
Suddenly there was a pause, and then a fresh outburst, this time
deliberate and controlled:
“Rah, rah rah! Rah, rah, rah! Rah, rah, rah! Hollister! Hollister!
Hollister!”
No need to tell him in so many words that the ball had gone over.
This was enough. They were cheering for him, and, as he opened
his eyes again, something like a mist came over them. Presently this
cleared away, and he found himself looking into Merriwell’s face.
“How are you feeling, old fellow?” the senior asked anxiously.
“Hurt any place? Or is it just wind you want?”
Hollister smiled.
“That’s all,” he said quickly. “Be all right in a minute.”
He hesitated for an instant.
“Say, Dick.”
Merriwell bent lower.
“Yes?” he questioned.
“I couldn’t help it, old man,” Bob said in a low tone. “I broke my
promise, and I reckon you must think me an awful rotter. I held out
as long as I could; but you needed me, Dick, and I couldn’t sit there
and see the fellows licked. But it’s the last time.”
“Do you really mean that, Bob?” Merriwell asked slowly. “Don’t you
think that the next game you see will tempt you just as you have
been tempted to-day?”
Hollister shook his head decidedly.
“No, sir!” he said emphatically. “I’m through. This is the last. I’ll be
content now to cut it out for good. I’ve shown what I could do, and
——”
Another thunderous burst of cheering came from the stands.
“Hollister! Hollister! Hollister!”
“Not even for that would I break my word to you again, Merriwell.
You believe me, don’t you, old fellow?”
For an instant Dick gazed keenly into the anxious eyes of his
friend. Then his face cleared and a smile curved the corners of his
mouth.
“Sure,” he said simply.

Thiele pioneer in statistics Steffen L. Lauritzen all chapter instant download

Uploaded by

Copyright:

Available Formats

Thiele pioneer in statistics Steffen L. Lauritzen all chapter instant download

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thiele pioneer in statistics Steffen L. Lauritzen all chapter instant download

Uploaded by

Copyright:

Available Formats

Download the full version of the ebook at

Thiele pioneer in statistics Steffen L.

Explore and download more ebook at https://ebookultra.com

Current Management of the Menopause Christian Lauritzen

Thinking Politics Perspectives in Ancient Modern and

Pioneer Jesuits in Northern Mexico Peter Masten Dunne

Pioneer 1st Edition H.Q. Mitchell

George Lohmann Pioneer Professional Keith Booth

Advances in Understanding Strategic Behaviour Game Theory

Computational Statistics Handbook with MATLAB 2nd Edition

French Advance and Other Lines 1st Edition Steffen

2 Adjustment of a series of observations of an instrument-constant

and proceed directly from the equation

or more conveniently (2.8)

in combination with y0 = z0.

or with more convenient calculation (2.11)

in combination with xn = yn.

of (2.3), then from the two ﬁrst of these equations

and thus according to (2.7) and (2.5)

with the single exception that (2.15)

the coefﬁcients in the equations (2.16)

will be given as (2.17)

One then obtains for the mean square errors

according to the relations

double application of (2.18) further gives

which is unchanged from (2.18). One thus has

consequently the result is that

3 Determination of the units for the two types of weights

Fig. 4 Geometric construction of the preceding adjustment

For the two groups of equations we ﬁnd in this fashion16

obtain for the determination of the units of weights (3.1)

etc. If then the unit of weights Ew is calculated as

4 Adjustment by linear functions of the unknown elements

moment to the next so that one has equations (4.1)

and in addition some more complex albeit analogous auxiliary variables

beginning from {r0s0} = r0s0.

leading to two equations for determining p and q(4.5)

and for {ra + 1sa + 1} we have

while using this to calculate {ra + 1sa + 1} as (4.9)

We ﬁnally note the formula

15 12.44 25.0 82.5 .0090 .23

THE BEGINNING OF THE GAME.

It must not be supposed that Bob Hollister’s course was an easy

A BROKEN PROMISE AND A VICTORY.

You might also like