Get the full ebook with Bonus Features for a Better Reading Experience on ebookmeta.

com

Fundamentals of Causal Inference with R 1st

Edition Babette A. Brumback

https://ebookmeta.com/product/fundamentals-of-causal-
inference-with-r-1st-edition-babette-a-brumback/

OR CLICK HERE

DOWLOAD NOW

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

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Causal Inference in Python: Applying Causal Inference in

the Tech Industry (Fourth Early Release) 4th Edition
Matheus Facure
https://ebookmeta.com/product/causal-inference-in-python-applying-
causal-inference-in-the-tech-industry-fourth-early-release-4th-
edition-matheus-facure/
ebookmeta.com

Causal Inference The Mixtape 1st Edition Scott Cunningham

https://ebookmeta.com/product/causal-inference-the-mixtape-1st-
edition-scott-cunningham/

ebookmeta.com

Causal Inference for Data Science MEAP Alex Ruiz De Villa

https://ebookmeta.com/product/causal-inference-for-data-science-meap-
alex-ruiz-de-villa/

ebookmeta.com

New Secrets The Conall Clan 6 1st Edition Donna Steele

Steele Donna

https://ebookmeta.com/product/new-secrets-the-conall-clan-6-1st-
edition-donna-steele-steele-donna/

ebookmeta.com
Wind and Dark Water Kildevil Cove Murder Mysteries 1
Guernsey 1 1st Edition J S Cook

https://ebookmeta.com/product/wind-and-dark-water-kildevil-cove-
murder-mysteries-1-guernsey-1-1st-edition-j-s-cook/

ebookmeta.com

Craft Beer Culture and Modern Medievalism Brewing Dissent

1st Edition Noëlle Phillips

https://ebookmeta.com/product/craft-beer-culture-and-modern-
medievalism-brewing-dissent-1st-edition-noelle-phillips/

ebookmeta.com

Paranormal Dating Agency 62 0 Flame it Up 1st Edition

Milly Taiden

https://ebookmeta.com/product/paranormal-dating-agency-62-0-flame-it-
up-1st-edition-milly-taiden/

ebookmeta.com

Public Relations and Conflict Management in Practice

Allieu Jalloh

https://ebookmeta.com/product/public-relations-and-conflict-
management-in-practice-allieu-jalloh/

ebookmeta.com

Right Where You Left Me 1st Edition Vanessa Stock

https://ebookmeta.com/product/right-where-you-left-me-1st-edition-
vanessa-stock/

ebookmeta.com
The Arsenal of Democracy Aircraft Supply and the Anglo
American Alliance 1938 1942 1st Edition Gavin J. Bailey

https://ebookmeta.com/product/the-arsenal-of-democracy-aircraft-
supply-and-the-anglo-american-alliance-1938-1942-1st-edition-gavin-j-
bailey/
ebookmeta.com
Fundamentals of
Causal Inference
with R
CHAPMAN & HALL/CRC
Texts in Statistical Science Series
Joseph K. Blitzstein, Harvard University, USA
Julian J. Faraway, University of Bath, UK
Martin Tanner, Northwestern University, USA
Jim Zidek, University of British Columbia, Canada

Recently Published Titles

Modern Data Science with R, Second Edition

Benjamin S. Baumer, Daniel T. Kaplan, and Nicholas J. Horton

Probability and Statistical Inference

From Basic Principles to Advanced Models
Miltiadis Mavrakakis and Jeremy Penzer

Bayesian Networks
With Examples in R, Second Edition
Marco Scutari and Jean-Baptiste Denis

Time Series
Modeling, Computation, and Inference, Second Edition
Raquel Prado, Marco A. R. Ferreira and Mike West

A First Course in Linear Model Theory, Second Edition

Nalini Ravishanker, Zhiyi Chi, Dipak K. Dey

Foundations of Statistics for Data Scientists

With R and Python
Alan Agresti and Maria Kateri

Fundamentals of Causal Inference

With R
Babette A. Brumback

Sampling
Design and Analysis, Third Edition
Sharon L. Lohr

Theory of Statistical Inference

Anthony Almudevar

Probability, Statistics, and Data

A Fresh Approach Using R
Darrin Speegle and Brain Claire

Bayesian Modeling and Computation in Python

Osvaldo A. Martin, Raviv Kumar and Junpeng Lao

Bayes Rules!
An Introduction to Applied Bayesian Modeling
Alicia Johnson, Miles Ott and Mine Dogucu

For more information about this series, please visit: https://www.crcpress.com/

Chapman--Hall/CRC-Texts-in-Statistical-Science/book-series/CHTEXSTASCI
Fundamentals of
Causal Inference
with R

Babette A. Brumback
First edition published 2022
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742

and by CRC Press

2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN

© 2022 Babette A. Brumback

CRC Press is an imprint of Taylor & Francis Group, LLC

The right of Babette A. Brumback to be identified as author of this work has been asserted by him/
her/them in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.

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

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

For permission to photocopy or use material electronically from this work, access www.copyright.
com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA
01923, 978-750-8400. For works that are not available on CCC please contact mpkbookspermis-
[email protected]

Trademark notice: Product or corporate names may be trademarks or registered trademarks and are
used only for identification and explanation without intent to infringe.

ISBN: 9780367705053 (hbk)

ISBN: 9780367705091 (pbk)
ISBN: 9781003146674 (ebk)

DOI: 10.1201/9781003146674

Typeset in CMR10
by KnowledgeWorks Global Ltd.

Visit the companion website/eResources: www.routledge.com/9780367705053

To my parents
Contents

Preface xi

1 Introduction 1
1.1 A Brief History . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Data Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Mortality Rates by Country . . . . . . . . . . . . . . . 6
1.2.2 National Center for Education Statistics . . . . . . . . 8
1.2.3 Reducing Alcohol Consumption . . . . . . . . . . . . . 9
1.2.3.1 The What-If? Study . . . . . . . . . . . . . . 9
1.2.3.2 The Double What-If? Study . . . . . . . . . 10
1.2.4 General Social Survey . . . . . . . . . . . . . . . . . . 13
1.2.5 A Cancer Clinical Trial . . . . . . . . . . . . . . . . . 15
1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 Conditional Probability and Expectation 19

2.1 Conditional Probability . . . . . . . . . . . . . . . . . . . . . 19
2.2 Conditional Expectation and the Law of Total Expectation . 21
2.3 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Sampling Distributions and the Bootstrap . . . . . . . . . . 29
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3 Potential Outcomes and the Fundamental Problem of Causal

Inference 37
3.1 Potential Outcomes and the Consistency Assumption . . . . 38
3.2 Circumventing the Fundamental Problem of Causal Inference 39
3.3 Effect Measures . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4 Effect-Measure Modification and Causal Interaction 59

4.1 Effect-Measure Modification and Statistical Interaction . . . 60
4.2 Qualitative Agreement of Effect Measures in Modification . . 71
4.3 Causal Interaction . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

vii
viii Contents

5 Causal Directed Acyclic Graphs 81

5.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

6 Adjusting for Confounding: Backdoor Method via

Standardization 99
6.1 Standardization via Outcome Modeling . . . . . . . . . . . . 99
6.1.1 Average Effect of Treatment on the Treated . . . . . . 105
6.1.2 Standardization with a Parametric Outcome Model . . 110
6.2 Standardization via Exposure Modeling . . . . . . . . . . . . 113
6.2.1 Average Effect of Treatment on the Treated . . . . . . 115
6.2.2 Standardization with a Parametric Exposure Model . 116
6.3 Doubly Robust Standardization . . . . . . . . . . . . . . . . 123
6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

7 Adjusting for Confounding: Difference-in-Differences

Estimators 133
7.1 Difference-in-Differences (DiD) Estimators with Linear,
Loglinear, and Logistic Models . . . . . . . . . . . . . . . . . 135
7.1.1 DiD Estimator with a Linear Model . . . . . . . . . . 136
7.1.2 DiD Estimator with a Loglinear Model . . . . . . . . . 137
7.1.3 DiD Estimator with a Logistic Model . . . . . . . . . 137
7.2 Comparison with Standardization . . . . . . . . . . . . . . . 138
7.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

8 Adjusting for Confounding: Front-Door Method 147

8.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
8.2 Theory and Method . . . . . . . . . . . . . . . . . . . . . . . 150
8.3 Simulated Example . . . . . . . . . . . . . . . . . . . . . . . 152
8.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

9 Adjusting for Confounding: Instrumental Variables 157

9.1 Complier Average Causal Effect and Principal Stratification 158
9.2 Average Effect of Treatment on the Treated and Structural
Nested Mean Models . . . . . . . . . . . . . . . . . . . . . . 160
9.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
9.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

10 Adjusting for Confounding: Propensity-Score Methods 175

10.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.2 Using the Propensity Score in the Outcome Model . . . . . . 179
10.3 Stratification on the Propensity Score . . . . . . . . . . . . . 180
10.4 Matching on the Propensity Score . . . . . . . . . . . . . . . 181
10.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Contents ix

11 Gaining Efficiency with Precision Variables 187

11.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
11.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
11.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

12 Mediation 195
12.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
12.2 Traditional Parametric Methods . . . . . . . . . . . . . . . . 199
12.3 More Examples . . . . . . . . . . . . . . . . . . . . . . . . . 200
12.4 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

13 Adjusting for Time-Dependent Confounding 209

13.1 Marginal Structural Models . . . . . . . . . . . . . . . . . . . 210
13.2 Structural Nested Mean Models . . . . . . . . . . . . . . . . 212
13.3 Optimal Dynamic Treatment Regimes . . . . . . . . . . . . . 215
13.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

Appendix 223

Bibliography 225

Index 233
Preface

In my experience, to learn a new statistical method, it can be very helpful to

reduce the problem as far as possible and work up from there. Often, I find it
easiest to understand a new method in the context of a ‘toy example.’ In this
spirit, whether learning or teaching statistical methods for causal inference,
I have found much success by focusing on what I will call a ‘binary dataset,’
that is, a dataset in which all variables are binary. This textbook represents
my effort to teach causal inference in that way. Most of the methods in the
book are introduced via examples in which all variables are either naturally
dichotomous or have been dichotomized by me. While I realize that this style
may not appeal to everyone, I am hopeful that it will aid students in under-
standing the methods in this book and in learning further statistical methods
that can be simplified in an analogous way. The book does not abandon more
complex datasets; for example, in the exercises for Chapters 3 and 6 students
learn about two-part models in the context of estimating conditional causal
effects and population-averaged causal effects via standaridzation.
Research in causal inference has exploded since my first encounter with it
in the late 1990s. I have chosen to focus only on fundamentals that could be
covered in a one-semester course. Students are encouraged to keep learning and
reaching out to the broader literature after the course ends. In keeping with
a focus on fundamentals, I have striven to limit the mathematical knowledge
required. I find it easiest to teach in terms of discrete probability distributions,
with the understanding that students who know calculus can easily substitute
integrals for sums. Chapter 2 reviews conditional probability and expectation,
introduces estimation via estimating equations, and reviews sampling distri-
butions and explains how to use the bootstrap. The mathematical level stays
there for most of the remainder of the book. However, there are many hard
concepts to learn, a lot of notation, and tough logical arguments throughout.
Therefore, I think it is fair to say that this is not an easy textbook, despite
my efforts to make it as easy as possible.
Thanks to the publisher and to the University of Florida, I had the
opportunity to pilot test the book with my course on Causal Inference in
Spring 2021. I reached out to undergraduates as well as graduate students,
and found myself with 38 very motivated students from B.S., M.S., and Ph.D.
programs in Statistics and Biostatistics and from Ph.D. programs in Educa-
tion and Epidemiology. I wanted the students to have exposure to all of the
exercises at the end of the chapters. Therefore, I put the students into groups
of 5–6 people and assigned them to work together on all of the exercises,

xi
xii Preface

dividing up the work of making a presentation for each one. Then, I randomly
called on students from the class to present their solutions. I made sure that
the groups were relatively balanced in terms of student levels. The pilot test
went well. The material is accessible to senior-level undergraduate students in
Statistics, and likely also in Data Science (the program at the University of
Florida was too new to have any of those). It is even more accessible to M.S.
students in Biostatistics and Statistics, and not too easy to avoid challenging
Ph.D. students from a variety of programs.
In my career, I have always been motivated by real examples. Therefore, I
worked hard to include as many real data examples as possible in the book and
in the exercises. I chose to analyze the data examples using R throughout the
book. Some familiarity with R is essential prior to beginning the textbook. The
appendix documents where to download R and which packages are necessary
for the R code presented in the book. Having been fortunate to teach very many
times out of the textbooks by J.J. Faraway that include R code and real data
examples, I thought his approach would also work well for causal inference.
Having said this, I also find it very helpful to learn and teach causal inference
methods using simulated data examples, for which the causal mechanisms are
known. Thus, I also make extensive use of simulations throught the text.
In very many of the examples, I do not really believe that the results can be
interpreted causally. Indeed, I am generally very cautious about interpreting
results causally, even when the required assumptions are more plausible. Nev-
ertheless, many methods in this book can also be applied when assumptions
do not hold, and in this case they have meaning as “adjusted associations,”
which are often still of interest to collaborators. Furthermore, I believe such
examples still have pedagogical value.
I have taught Causal Inference three times. First, in Fall of 1999 at the
University of Washington, when I was just starting out and did not know
much at all; furthermore, textbooks on the topic were virtually non-existent.
Second, in Spring of 2020 at the University of Florida, when I taught out of
the new textbook by M.A. Hernan and J.M. Robins. This motivated me to try
to present the material in my own way when I taught for the third time and
piloted this textbook in Spring of 2021 at the University of Florida. I wish to
thank all of my previous mentors and especially the students of those three
classes, who asked many good questions that taught me a lot. Students of the
third class pointed out many typos and errors in the textbook and also made
suggestions for how to improve it. I am grateful to all, but wish to specifically
acknowledge those who gave me written permission: Seungjun Ahn, Saurabh
Bhandari, James Colee, Amy M. Crisp, Deborah Rozum, Jeremy Sanchez,
Jake Shannin, Eric A. Wright, Dongyuan Wu, Xiulin Xie, Kai Yang, Xiaomeng
Yuan, Wenjie Zeng, Runzhi Zhang, and Xinyi Zhang. Many thanks also to
Edward Kennedy, Nandita Mitra, and Ronghui Xu for their helpful reviews.
A website for the book with datasets, R code, solutions to the
odd exercises, and more is available by searching for the textbook at
www.routledge.com.
1
Introduction

1.1 A Brief History

The history of causal inference is long and complex, with many thinkers from
a variety of disciplines having written on the topic. A good overview from
a statistician’s perspective is provided by Holland (1986). Here, we briefly
touch upon some prominent milestones in the field. Our earliest stop is from
Aristotle (350BC), who enumerated four types of causes.
1. What a thing is made of, e.g. the porcelain of a cup
2. The form of a thing, e.g. the octave the relation of 2:1
3. “The primary source of the change or coming to rest; e.g. the man
who gave advice is a cause, the father is cause of the child, and
generally what makes of what is made and what causes change of
what is changed.”
4. “That for the sake of which a thing is done, e.g. health is the cause of
walking about. (‘Why is he walking about?’ we say. ‘To be healthy,’
and, having said that, we think we have assigned the cause.)”
In statistics, causes are of Aristotle’s type 3, “what causes change of what is
changed.” We are commonly interested in the effect of an intervention, or the
cause of a health or societal problem. Type 4 causes are interesting to ponder,
as these might be taken to imply that the future can cause the past, in a certain
sense. However, if we translate the definition so that our intention to achieve
the end is the cause, then our intention (to be healthy) can be posited to exist
prior to, and thus be a type 3 cause of, the action (walking) which in turn
causes (again in the sense of type 3) the end (health). If divine intervention
were possible, it could be that the intention belonged to a supreme being,
who then manipulated the sequence of events at least as often as necessary
to reach the desired end, or fate. In that semi-deterministic or deterministic
world, whether the future caused the past or the past caused the future would
be irrelevant. We would be watching a movie directed by the supreme being,
needlessly concerning ourselves with optimal decision-making.
Hume (1738) famously wrote in 1738 that cause and effect is not much
more than the constant conjunction of events.

DOI: 10.1201/9781003146674-1 1
2 Introduction

“We have no other notion of cause and effect, but that of certain objects,
which have been always conjoined together, and which in all past instances
have been found inseparable. We cannot penetrate into the reason of the
conjunction. We only observe the thing itself, and always find that, from
the constant conjunction, the objects require a union in the imagination...
Thus, though causation be a philosophical relation, as implying contiguity,
succession, and constant conjunction, yet it is only so far as it is a natural
relation, and produces a union among our ideas, that we are able to reason
upon it, or draw any inference from it.”
It is noteworthy that Hume did not propose any methods to verify a causal
relationship other than to observe contiguity, succession, and constant con-
junction. Nevertheless, shortly afterwards in 1747, James Lind undertook what
is widely recognized today as one of the very first medical trials, proving a
causal relationship between eating oranges and lemon and recovering from
scurvy. From Brown (2005), we read that Lind took aside twelve men with
advanced symptoms of scurvy “as similar as I could have them.” Six pairs of
men aboard the HMS Salisbury were thus experimented upon. The first pair
were given slightly alcoholic cider. The second were given an elixir of vitriol.
The third pair took vinegar. The fourth drank sea water. The fifth were fed
two oranges and one lemon daily for six days, when the meager supply ran
out. The sixth were given a medicinal paste and cream of tartar, which is a
mild laxative. The pair who were fed the oranges and the lemons were nearly
recovered after only a week. Those who had drunk the cider responded favor-
ably, but at the end of two weeks they were still too weak to return to duty.
The other four pairs did not experience good effects.
By 1846, Mill (1846), in his five canons, proposed five methods for proving
cause and effect:
“First Canon (the Method of Agreement): If two or more instances of the
phenomenon under investigation have only one circumstance in common,
the circumstance in which alone all the instances agree, is the cause (or
effect) of the given phenomenon.
Second Canon (the Method of Difference): If an instance in which the
phenomenon-under investigation occurs, and an instance in which it does
not occur, have every circumstance save one in common, that one occurring
only in the former; the circumstance in which alone the two instances differ,
is the effect, or cause, or a necessary part off the cause, off the phenomenon.
Third Canon (the Joint Method of Agreement and Difference): If two or
more instances in which the phenomenon occurs have only one circum-
stance in common, while two or more instances in which it does not occur
have nothing in common save the absence of that circumstance; the cir-
cumstance in which alone the two sets of instances differ, is the effect, or
cause, or a necessary part of the cause, of the phenomenon.
Discovering Diverse Content Through
Random Scribd Documents
expectation that the particle will permanently adhere to the surface of
the filter.

Testing Smoke Filters

All the early tests made on smoke filters used
diphenylchloroarsine, because it was felt that the filter must be
tested against a toxic smoke. A man test was developed as
representative as possible of actual conditions in the field, and the
time necessary for a man to detect diphenylchloroarsine smoke in
the effluent stream when breathing at a normal rate, using a carefully
controlled concentration of smoke produced by detonation, was used
as the criterion of the protection offered by the canister. This test was
subject to extensive individual variations, due to the varying
physiological resistances of different men to diphenylchloroarsine
smoke. Further, it was quite inadequate for rapid testing on a large
scale. A testing machine was then developed, which gave results
comparable with those obtained in the man test. The method used in
detecting the gas was physiological, that is, by smell or by its
irritating action towards the membranes of the eye. While these are
purely qualitative tests, they are much more sensitive than any
possible chemical tests.
Because of the desirability of having a method which could be
controlled chemically, other methods were developed.
Ammonium chloride is a solid smoke, consisting of particles of
quite variable sizes. It is sensitive to dilution and clogs the pores of
the filtering medium quite rapidly. For this reason it was used in the
study of the rate of plugging or clogging of the filter (the closing of
the pores of the fabric or other material to the passage of air).
The smoke is produced by the reaction of ammonia and
hydrogen chloride-air streams. The smoke thus generated is passed
from the mixing chamber to a larger distribution box and from there
through the filter, at a standard rate. The concentration of the smoke
may be accurately determined by chemical means or
photometrically, using a Hess-Ives Tint Photometer, the Marten
Photometer, or a special photometer developed by the Chemical
Warfare Service.
A comparison of a large number of tests with those of other
smokes would indicate that ammonium chloride smoke offers
accurate information as to protection sought, but is hardly a
desirable smoke for testing on a large scale.
The third method developed was the sulfuric acid smoke. This
smoke was produced by passing dry air through a tower of solid
pieces of sulfur trioxide and then mixing the vapor with a large
volume of air at 50 per cent relative humidity. It is not a clogging
smoke and the filtering efficiency does not change materially in the
time of exposure required for a test. The smoke lends itself easily to
chemical analysis and offers data as to exact particulate cloud
concentrations which will penetrate canisters; photometric
measurements are also applicable.
 Fig. 106.—Tobacco Smoke Apparatus for Testing
Canisters.

The fourth method consists in the use of tobacco smoke. This is

generated by passing air over ignited sticks of a mixture of tobacco
(63 per cent), rosin (30 per cent) and potassium nitrate (7 per cent).
This smoke is composed of particles of extreme uniformity in size;
chemically it is relatively inert. It is not a clogging smoke and is not
sensitive to moisture and dilution. The density of the effluent smoke
is compared with that of the entering smoke in a Tyndall beam, and
the filtering capacity of the material determined in terms of the
amount of air necessary to dilute the entering air to the same
concentration of the effluent air. The method is simple in
manipulation and the test is a rapid one (50 canisters per day).
Because of the apparent superiority of tobacco smoke as a testing
smoke, the accompanying disadvantages are possibly outweighed.
 From the standpoint of inherent chemical properties, the general
desirability of a suitable testing smoke would decrease in the
following order: tobacco, sulfuric acid, ammonium chloride.
 CHAPTER XIX
SIGNAL SMOKES

The success of pyrotechnics in night signalling led, during the

World War, to considerable attention being paid to the development
of pyrotechnic signals for day use. This was mainly directed to the
production of distinctive smokes, which should have the same long
range visibility under varying light conditions. Since a gray or white
smoke might be confused with the smoke produced accidentally by
the explosion of shell, it was necessary to use smoke of definite and
unmistakable colors, and red, blue, yellow, green and purple smokes
were developed. During the early part of the war, only a yellow
smoke was in use, though others were added later.

Production of Colored Smokes

There are three possible ways of obtaining signal smokes.

I. Mechanically dispersing solids.

II. Chemical Reaction.
III. Volatilization of colored materials.

I. The first method, while possible, can never be an efficient

method of producing signals. Some success was met with in
dispersing certain inorganic materials, as rouge, and ultramarine, in
projectiles fired from a 3-inch mortar and exploded by a time fuse
arrangement at the height of their flight. Various mixtures were also
tried, such as antimony oxysulfide and aluminum powder (red),
arsenic and antimony trichlorides with sodium thiosulfate (yellow),
etc., but these compositions have the disadvantages of being liable
to catch fire if dispersed by a black powder explosion.
 II. While colored smokes may be produced by chemical reaction,
such as the combination of hydrogen iodide (HI), chlorine and
ammonia, the clouds are not satisfactory as signals. In this particular
case, the purple cloud (to the operator in the aeroplane) appeared
white to the observers on the ground.
High temperature combustion smokes have also been studied.
These are used in the so-called smoke torches. The yellow arsenic
sulfide smoke is the most widely used. Most formulas call for some
sulfide of arsenic (usually the native realgar, known commercially as
“Red Saxony Arsenic”), sulfur, potassium nitrate, and in some cases,
a diluent like ground glass or sand. A typical mixture consists of:
Red arsenic sulfide 55%
Sulfur 15%
Potassium nitrate 30%
A very similar smoke may be obtained from the following mixture:
Sulfur 28.6%
White arsenic 32.0%
Potassium nitrate 33.8%
Powdered glass 6.6%
These smokes are not as satisfactory in color as the smoke
produced by a dye smoke mixture, especially when viewed from a
distance, with the sky as a background. They fade out rather quickly
to a very nearly white smoke.
A black smoke upon first thought might seem to be the easiest of
all smokes to produce, but actually the production of a black smoke
that would be satisfactory for signalling purposes was rather a
difficult matter.
Starting with the standard smoke mixture, which gives a white or
gray smoke, hexachloroethane, which is solid, was substituted for
the carbon tetrachloride, in order to avoid a liquid constituent.
Naphthalene was first used, until it was found that the mixture of
naphthalene and hexachloroethane melted at temperatures below
that of either of the constituents. Anthracene was then substituted.
The principal reaction is between the magnesium and the chlorine-
containing compound, whereby magnesium chloride and carbon are
formed. The reaction is very violent, and a white smoke is produced.
The anthracene slows down the reaction and at the same time colors
the smoke black. The speed of the reaction may be controlled by
varying the anthracene content.
In burning this type of smoke mixture in a cylinder, it is essential
that free burning be allowed. It has been found that if combustion is
at all smothered, and the smoke forced to escape through a
comparatively small opening, it will be gray instead of dense black.
III. Various attempts have been made to utilize the heat evolved
when the Berger type smoke mixture reacts to volatilize or
mechanically disperse various colored inorganic substances, and
especially iodine. These were unsuccessful. Modifications, such as
Strontium nitrate 1 part
Powdered iron 2 parts
Iodine 3 parts
were also tried, but while such mixtures ignited easily, burned freely
and evenly, and gave a continuous heavy purple cloud, they were
very sensitive to moisture and capable of spontaneous ignition.
The most satisfactory and successful colored smokes are those
produced by the volatilization of organic dye materials. This practice
seems to have originated with the British, who produced such
smokes by volatilizing or vaporizing special dyes by igniting mixtures
of the dye, lactose and potassium chlorate and smothering the
combustion.
In selecting dyestuffs for this purpose it was at once recognized
that only those compounds can be used which are volatilized or
vaporized without decomposition by the heat generated when the
mixture is ignited and the combustion smothered. It was also found
that the boiling point and melting or volatilization point of the colored
compound must be close enough together so that there is never
much liquid dye present. Since all colored organic compounds are
destroyed if subjected to sufficient heat, the mixture must be so
prepared and the ignition so arranged that the heat generated is not
sufficient to cause this destruction.
The oxidizing agents used in the combustion mixture may be
either potassium or sodium chlorate. The nitrate is not satisfactory.
Lactose has proven the best combustible. Powdered orange shellac
is fairly satisfactory but offers no advantage over lactose.
The following dyes have been found to give the best smokes:
Red “Paratoner”
Yellow Chrysoidine + Auramine
Blue Indigo
Purple Indulin (?)
Green Auramine Yellow + Indigo
At the beginning of the war, the only colored smoke used by the
United States Army was a yellow smoke. The smoke mixture used in
all signals, excepting the smoke torch, was the old arsenic sulfide
mixture. The following smoke signals were adopted during the World
War:
Yellow and
Signal Parachute Rocket
Red
V. B. Parachute Cartridge Yellow
25 mm. Very Parachute Yellow
Cartridge
35 mm. Signal Cartridge Yellow
35 mm. Signal Cartridge Red
35 mm. Signal Pistol
25 mm. Very Signal Pistol
V. B. Rifle Discharger Cut
 The Tactical Use of Signal Smokes
From the days when Horatius kept the bridge, down through the
centuries to the World War, all leaders in battle were pictured at the
front and with flaming sword, mounted on magnificent chargers, or
otherwise so prominently dressed that all the world knew they were
the leaders. During all these hundreds of years commands on the
field of battle were by the voice, by the bugle, or by short range
signals with arms, flags, and swords. Even where quite large forces
were involved they were massed close enough ordinarily so that
signalling by such means sufficed to cover the front of battle. In
those cases where they did not, reliance was put upon swift couriers
on horseback or on foot.
With the invention of smokeless powder and the rifled gun battles
were begun and carried on at greater and greater ranges. Artillery
fired not only 2,000 to 3,000 yards but up to 5,000 and 10,000 yards,
or even, as in the World War, at 20,000 yards and more. It was then
that other means of signalling became essential. Distant signalling
with flags is known to have been practiced to a certain extent on land
for a long time. The extension of the telegraph and telephone
through insulated wires laid by the Signal Service was the next great
step in advance, and in the World War there came in addition the
wireless telephone both on land and in aeroplanes and balloons.
Along with this development, as mentioned under Screening
Smokes, came the development of the use of smoke for protection
and for cutting off the view of observers, thus making observation
more and more difficult. This use of smoke, coupled with the deadly
fire of machine guns and high explosives, forced men to take shelter
in deep shell holes, in deep trenches and other places that were
safe, but which made it nearly impossible to see signals along the
front of battle.
Every man can readily be taught to read a few signals when
clearly indicated by definite, sharply defined colored smokes. At first
these were designed for use on the ground and will be used to a
certain extent in the future for that purpose, particularly when it is
desired to attract the attention of observers in aeroplanes or
balloons. In such cases a considerable volume of smoke is desired.
For the man in the trench or shell hole some means of getting the
signal above the dust and smoke of the battlefield is needed. It is
there that signal smokes carried by small parachutes, contained in
rockets or bombs, have proven their worth. These signals floating
high above the battlefield for a minute or more, giving off brilliantly
colored smokes, afford a means of sending signals to soldiers in the
dust and smoke of battle not afforded by any other method so far
invented. As before stated, every man can be taught these simple
signals, where but very few men can be taught to handle even the
simplest of wireless telephones.
Thus, smoke has already begun to complicate, and in the future
will complicate still more, every phase of fighting. It will be used for
deception, for concealment, for obscuring vision, for signalling and to
hide deadly gases. The signal rocket will be used to start battles,
change fronts, order up reserves, and finally to stop fighting.
The signal smokes by day will be displaced at night by brilliantly
colored lights which will have the same meaning as similarly colored
smokes during the day. Thus, literally, smoke in the future will be the
cloud by day and the pillar of fire by night to guide the bewildered
soldier on the field of battle with all its terrors and amidst the
confusion, gas, smoke and dust that will never be absent while
battles last.
 CHAPTER XX
INCENDIARY MATERIALS

Since it is generally known that white phosphorus, when exposed

to the air, takes fire spontaneously, it logically follows that numerous
suggestions should have been made for using this material in
incendiary devices. Practice, however, has shown that, while
phosphorus is undoubtedly of value against very easily ignitable
materials, such as hydrogen in Zeppelins, or the gasoline tanks of
aeroplanes and dry brush or grass, it is of much less value when
wood and other materials are considered. This is partly because of
the low temperature of burning, and partly because the product of
combustion (phosphoric anhydride) is really an excellent fireproofing
substance. In view of this, phosphorus was used primarily for smoke
production.
A superior incendiary material is found in thermit, a mixture of
aluminum and iron oxide. When ignited, it produces an enormous
amount of heat very quickly, and the molten slag that results from the
reaction will prolong the incendiary action upon inflammable
materials. When used alone, however, it has the disadvantages that
the incendiary action is confined to a small area and that the heat
energy is wasted because of the fact that it is so rapid in its action.
For this reason it is customary to add a highly inflammable
material, which will become ignited by the thermit and will continue
the conflagration. Petroleum oils, carbon disulfide, wood distillation
products and other inflammable liquids were thoroughly tested for
this purpose. The final conclusion was reached that oil, solidified with
soap (sodium salts of the higher fatty acids) by a special method
developed by the Chemical Warfare Service, was by far the best
material to be used. In certain tests, using a combination of thermit
and solid oil, flames fifteen feet high were obtained, which would be
very useful against walls, ceilings, etc.
 In addition to this type of incendiary material, it was desirable to
have a spontaneously inflammable mixture of oils, which could be
used in Livens’ shell, Stokes’ shell or aeroplane bombs. The basis of
these mixtures is fuel oil and phosphorus. By varying the proportions
of the constituents it is possible to obtain a mixture that will ignite
immediately upon exposure to the air, or one that will have a delayed
action of from 30 seconds to two minutes.
The incorporation of metallic sodium gives a mixture that will
ignite when spread upon water surfaces.

Incendiary Devices
The incendiary devices used during the late war included: bombs,
shell, tracer shell and bullets, grenades, and flame throwers.

Bombs

Incendiary bombs were used almost exclusively by aircraft. The

value of bombs which would cause destruction by starting
conflagrations was early recognized but their development was
rather slow. While the designs were constantly changing, two stand
out as the most favored: a small unit, such as the Baby Incendiary
Bomb of the English, and a large bomb, such as the French Chenard
bomb or the American Mark II bomb.
In general bombs which, when they function upon impact, scatter
small burning units over a considerable area, are not favored. Small
unit bombs can be more effectively used because the scatter can be
better regulated and the incendiary units can be more
advantageously placed.
German Bombs. Incendiary bombs were used by the Germans
in their airplane raids, usually in connection with high explosive
bombs. A typical armament of the later series of German naval
airships consisted of the following:
2 660-pound bombs
 10 220-pound bombs
15 110-pound bombs
20 Incendiary bombs
making a total weight of about 2½ tons.

Fig. 107.—Incendiary Devices.

(From Left to Right).
Mark II Bomb, B. I. Bomb, Mark I Dart, Mark II Dart,
Mark I Dart, Grenade, Mark I Bomb.

A typical German bomb is shown in Fig. 108. It consists

essentially of a receiver of white iron (r) composed of a casing and a
central tube of zinc, joined together in such a fashion that, when the
whole was complete, it had the appearance of an elongated vessel
with a hollow center. Within this central hollow is placed a priming
tube (t) of thin sheet iron, pierced by a number of circular openings.
The receiver is about 445 mm. (17.5 in.) high and 110 mm. (4.3 in.)
at its maximum diameter. It is wrapped with strands of tarred cord
over nearly its entire length. The empennage (270 mm. or 10.6 in.—
in height) consisted of three inclined balancing fins, which assured
the rotation of the projectile during its fall.
In the body of the bomb was a viscous mass of benzine
hydrocarbons, while the lower part of the receiver contained a
mixture of potassium perchlorate and paraffin. The central tube
apparently contained a mixture of aluminum and sulfur.
Fig. 108.—Aerial Incendiary Bomb,
November, 1916.
 Fig. 109.—German Incendiary Bomb,
Scatter Type.

All dimensions in millimeters.

Later the Germans used a scatter type of bomb (Fig. 109) which
was designed to give 46 points of conflagration. Each of these 46
small cylinders contained 50 grams of an air incendiary material.
They were arranged in layers, packed in with very fine gun powder.
The bomb is ignited by a friction lighter which is pulled automatically
when the bomb is released from the aeroplane. The bomb is
constructed to burst in the air and not on striking the ground. The
upper part of the projectile consists of a cast iron nose riveted to the
sheet iron body of the bomb. When the explosion occurs, the nose is
blown away and the small incendiary cylinders are scattered in the
air.
The incendiary material appears to be a mixture of barium nitrate
and tar. Its incendiary power is very low because combustion takes
the form of a small flame of very short duration. It should, however,
be very valuable for firing inflammable materials.
British Bombs. The early British bombs were petrol bombs,
which were used without great success for crop burning. Phosphorus
bombs were then used for attacking aircraft. But the most successful
incendiary is the so-called “Baby Incendiary Bomb.” This is a 6.5-
ounce bomb with an incendiary charge of special thermit. These
small bombs are carried in containers holding either 144 or 272
bombs. The former container approximates in size and weight one
50-pound H.E. bomb and the latter one 120-pound H.E. bomb. The
bomb contains a cartridge very much like a shot gun shell which, on
impact, sets down on the striker point in the base of the body, and
causes the ignition of the charge. It is claimed that the cartridge of
the B.I. bomb burns when totally immersed in any liquid (water
included) and in depths up to two feet the flame breaks through the
surface.
French Bombs. The French used three types of incendiary
bombs, a special thermit (calonite), the Chenard and the Davidsen.
The Chenard bomb is a true intensive type and is thought to be very
successful. It functions by means of a time fuse operated by the
unscrewing of a propeller, before striking the ground, and reaches its
target in flames. Its chief disadvantage is the small amount of
incendiary material which it carries. The Davidsen bomb expels its
charge as a single unit and is not considered as valuable or as
successful as the Chenard.
American Bombs. The program of the Chemical Warfare
Service included three types of bombs:
Mark II Incendiary drop bomb
Mark III Incendiary drop bomb
Mark I Scatter bomb
Mark II Bomb. The incendiary Mark II drop bomb is designed to
be dropped from an aeroplane and is intended for use against
buildings, etc., when penetrating effect followed by an intensive
incendiary action is sought.
The bomb case consists of two parts: a body and a nose. The
body is a tapering zinc shell which carries the firing mechanism and
stabilizing tail fin at the small end and at the large end a threaded
ring which screws into the nose. The nose is of drawn steel of such
shape as to have low end-on resistance and is sufficiently strong to
penetrate frame structures.

Fig. 110.—Loading Bombs with “Solid” Oil.

The incendiary effect is produced by a thermit charge carried in

the nose of the bomb. This charge is ignited by a booster of “Thermit
Igniter” fired by black powder. The latter is ignited by a flash from the
discharge of a standard 0.30 caliber service cartridge contained in
the body of the bomb, and exploded by a firing mechanism of the
impact type. This method of firing has proven wholly unsatisfactory
and will be superseded by some more direct-acting mechanism. The
body of the bomb is filled with solidified oil. The molten thermit burns
through the case of the bomb and liberates the oil which has been
partially liquefied by the heat of the thermit reaction. Additional
incendiary effect is afforded by the sodium nitrate contained in the
nose below the thermit, and by two sheet lead cylinders filled with
sodium and imbedded in the solid oil. The sodium increases the
difficulty of extinguishing the fire with water.
Mark III Bomb. This bomb is simply a larger size of the Mark II
bomb, its weight being approximately 100 pounds as compared with
40 pounds for the Mark II bomb. It is designed to be dropped from an
aeroplane and is intended for use against buildings when marked
penetrating effect is desired. The method of functioning is the same
as the Mark II bomb and it has the same defects in the firing
mechanism.
Mark I Bomb (Scatter Type). The Mark I incendiary drop bomb
is also designed to be dropped from aeroplanes and is intended for
use against grain fields, ammunition dumps, light structures or
similar objectives when only a low degree of ignition is required. It is
of the so-called scatter type, due to the action of the exploding
charge which casts out incendiary material within a radius of 20 feet
from the point of contact.
The incendiary action is due to the ejection of the various
incendiary units in the bomb by the explosion of the black powder in
the nose. The flash of this explosion serves to ignite the units. A
powder charge in the rear of the bomb acts simultaneously with the
nose charge, opening the bomb casing, and aiding materially in the
scatter of the units. The bomb is so arranged as to function close to
the ground, which is a further factor in the scatter of the units.
The incendiary units are waste balls about 2.5 in. in diameter and
having an average weight of 2.5 ounces, tied securely with strong
twine. These are soaked in a special oil mixture. Carbon disulfide
and crude turpentine, or carbon disulfide, benzene heads and crude
kerosene gave satisfactory results. A later development attempted to
replace the waste balls by solid oil, but the difficulties of manufacture
and questions of transportation argued against its adoption.
These bombs were not used at the front. Nearly all of the
American incendiary bombs proved too light on the nose and lower
half, generally resulting in deformation upon impact and very poor
results. New ones will be made stronger.

Incendiary Darts

The British early recognized the value of a small bomb, and

consequently adopted their B.I.B. (Baby Incendiary Bomb), weighing
about 6.5 ounces. These are capable of being dropped in lots of 100
or more and thus literally shower a given territory with fire. The
intensity of fire at any given point is much less than that obtained
with the larger bombs, but the increased area under bombardment
more than counter balances this disadvantage. While the British
aimed at the perfection of a universal bomb, the American service
felt that two classes should be developed, one to be used against
grain fields and forests, the other against buildings.
The first class was called the Mark I Dart. This consisted of an
elongated 12-gauge shot gun shell, filled with incendiary material
and provided with a firing mechanism to ignite the primer as the dart
strikes the ground. The flash of the primer sets fire to the booster,
which, in turn, ignites the main incendiary charge. The latter burns
several minutes, with a long flame. A retarding stabilizer attached to
the tail of the dart serves the two-fold purpose of insuring the
functioning of the firing mechanism and, by retarding the final
velocity of the dart, preventing the collapse of the dart body when
dropped from very high altitudes.
The incendiary mixture is one which gives a long hot flame, burns
for several minutes and leaves very little ash. In general it consists of
an oxidizing agent (barium or sodium chlorate), a reducing agent
(aluminum, or a mixture of iron, aluminum and magnesium), a filler
(rosin, powdered asphaltum or naphthalene) and in some cases a
binder (asphaltum, varnish or boiled linseed oil).
The Mark II Dart was developed to furnish a small size
penetrating agent. It consists of a two-inch (diameter) zinc case filled
with thermit and solid oil as the incendiary materials and provided
with a cast iron nose for penetration. During the first half minute after