David J. Smith - Basic Statistical Techniques For Medical and Other Professionals - A Course in Statistics To Assist in Interpreting Numerical Data-Productivity Press (2021)

Basic Statistical
Techniques for Medical

and Other
Professionals
Basic Statistical
Techniques for Medical
and Other
Professionals
A Course in Statistics to
Assist in Interpreting
Numerical Data
Dr. David J. Smith

Foreword by Sam Samuel
First Published 2022
by Routledge
605 Third Avenue, New York, NY 10158
and by Routledge
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
Routledge is an imprint of the Taylor & Francis Group, an Informa business
© 2022 David J. Smith
The right of David J. Smith to be identified as author[/s] of this work has been
asserted by him in accordance with sections 77 and 78 of the Copyright, Designs
and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilized
in any form or by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying and recording, or in any information
storage or retrieval system, without permission in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered
trademarks and are used only for identification and explanation without intent to
infringe.
Library of Congress Cataloging-in-Publication Data
A catalog record for this title has been requested
ISBN: 978-1-032-11495-8 (hbk)

ISBN: 978-1-032-11494-1 (pbk)
ISBN: 978-1-003-22013-8 (ebk)
DOI: 10.4324/9781003220138
Typeset in Garamond
by MPS Limited, Dehradun
Contents
Foreword ................................................................ix
Preface....................................................................xi
Acknowledgements ............................................. xiii
About the Author .................................................. xv
Introduction ........................................................xvii
1 Why One Needs Statistical Techniques .............1

Variables and Attributes .............................................. 3
Spread of a Variable.................................................... 3
Correlation ................................................................... 4
Taking Samples ........................................................... 4
Very Large and Very Small Numbers ......................... 5
2 Probability and Its Rules ..................................7
Empirical versus À Priori ............................................ 7
Combining Probabilities.............................................. 8
Mutually Exclusive Events ........................................ 11
Conditional Probabilities........................................... 13
Given a Positive Test ................................................ 13
Given a Negative Test............................................... 14
Comparison of the Two Outcomes.......................... 15
v
vi ▪ Contents
3 Dealing with Variables....................................17

Metrics........................................................................ 17
The Mean ................................................................... 19
The Standard Deviation (and Variance)................... 19
Grouped Data............................................................ 20
A Range for Each Value............................................ 20
Making Inferences from the Mean
and Standard Deviation ............................................ 20
The Coefficient of Variation...................................... 23
Other Measures of Central Tendency
(Types of Average).................................................... 25
Medians...................................................................... 25
Geometric Mean ........................................................ 27
4 Comparing Variables .......................................29
Comparing the Standard Deviations ........................ 29
Comparing the Means ............................................... 31
Double- and Single-Sided Inferences....................... 32
5 Presenting Data and Establishing
Trends .............................................................33
Visual Presentation of Data ...................................... 33
Histograms ................................................................. 33
Graphs........................................................................ 38
Some Pitfalls .............................................................. 41
Supression.................................................................. 41
Extrapolation.............................................................. 42
Logarithmic Scales ..................................................... 43
Moving Averages ....................................................... 45
Control Charts............................................................ 47
Contents ▪ vii
6 Dealing with Attributes...................................49

Binomial..................................................................... 49
Poisson....................................................................... 51
7 Testing for Significance (Attributes) ...............55
8 Correlation and Regression.............................61
Relating Two Variables ............................................. 61
False Correlation........................................................ 65
Establishing a Normal Distribution........................... 66
9 Handling Numbers (Large and Small).............69
Big Numbers.............................................................. 69
Small Numbers .......................................................... 70
Multiplying and Dividing .......................................... 71
Some Examples ......................................................... 72
Familiarity with Negative Exponents........................ 72
10 An Introduction to Risk ..................................75
Individual Risk of Fatality (per Exposure to
an Activity)................................................................. 75
Individual Risk of Fatality (per Annum) .................. 77
Fatal Accident Frequency (FAFR)............................. 78
Maximum Tolerable Risk .......................................... 78
As Low as Reasonably Practicable (ALARP) ............ 80
Variable versus Constant Rates................................. 82
11 A Final Word ...................................................83
Appendices ............................................................85
Index ...................................................................113
Foreword
I have had the pleasure of knowing David Smith for nearly

40 years. He has run a successful consultancy business
involving Reliability and Risk throughout that time, which is
heavily dependent on statistical analysis.
The idea for this book arose because a relative, having
attained a degree in Neuroscience, was undertaking an MSc to
further her career, which has a significant statistical element.
David initially produced training material to provide the basic
skills required. During the course, the material was continually
updated, to the point where he considered the exercises and
slides worth converting into a textbook. My conclusion is that
it was the right decision.
Although the book is aimed specifically at members of the
medical fraternity (including students), it concentrates on
practical analysis methods rather than the underlying theory.
It is a compromise between simple practical methods and
unnecessary mathematics and should be equally valuable to
anyone taking a high-level degree course that involves
ix
x ▪ Foreword
knowledge of basic statistical techniques. It is also aimed at

those already qualified and in practice in most professions
where numerical data is involved.
Sam Samuel
Preface
The purpose of this book is to provide members of the

medical and other professions, including scientists and
engineers, with a basic understanding of statistics and
probability. It does not seek to confuse the reader with in-
depth mathematics but provides basic methods for
interpreting data and making inferences. Although the
worked examples have a medical flavour the principles
apply to the analysis of any numerical data.
The information in this book is for guidance and is not
intended to give specific instructions to professionals, with
whom the sole responsibility for patient/client safety lies.
No responsibility is assumed by the author/publisher for
any consequences arising from actions taken on the basis
of the material herein.
xi
Acknowledgements
My sincere thanks to Sam Samuel, without whose patience

and help I would not have completed this little book. He
has been a life-long colleague, associate and friend and is
a highly respected consultant in the specialised field of
quantified risk assessment. His very major contributions to
the text and material in this book have been invaluable.
xiii
About the Author
Dr David J Smith BSc, PhD, FIET, FCGI, HonFSaRS,

MIGEM, is the Proprietor of Technis Consultancy. He has
written numerous books on Reliability, Risk and Statistics
during the last 40 years. He has also written a range of articles
(and a book) on the statistics of helicopter safety. David is the
past President of the UK Safety and Reliability Society.
By the same author:

Reliability Maintainability and Risk, 9th Edition, Smith
DJ, Elsevier ISBN 9780081020104.
The Safety Critical Systems Handbook 5th edition, Smith
DJ and Simpson KGL, (Elsevier) ISBN 9780128207000.
xv
Introduction
We are surrounded by numerical data in every walk of life.

Medical professionals record blood chemistries, blood
pressures, blood sugars, response times, optical measure-
ments, pain and quality-of-life-related metrics and so on.
The list is endless. Engineers and scientists practice only by
virtue of recording and making judgements from numbers,
which record the physical parameters that describe our
world. Every profession involves some aspect of counting
something, be it money, persons, times, market responses,
patterns of criminality and so on.
There are numerous excellent textbooks on the subject of
statistics that explain the mathematics in detail and provide
a whole range of sophisticated techniques for analysing
numerical data. Those embarking on a career of scientific
research (medical or otherwise) need to study the subject at
that level. However, the vast majority of professionals do
not require that depth, and this book is intended to satisfy
the need for a relatively short course which explains and
illustrates simple, but nevertheless powerful, techniques for
making important judgements that would otherwise not be
arrived at.
xvii
xviii ▪ Introduction
I may be accused of oversimplifying the subject but I am

convinced that the techniques described here are more than
adequate for the vast majority of people.
Statistics has been called, amongst other things, an inexact
science and is often regarded with suspicion by the unini-
tiated. The subject is surrounded by misunderstanding and,
furthermore, the distinction between probability and
statistics is not always clear. We shall, therefore, begin by
looking at the meaning of the two terms.
Statistics is simply the science of interpreting numerical
data. If, in a certain brand of matches, 99.9% fall within the
range 4.8 cm, plus or minus 0.3 cm, then it could be
inferred that a match 5.25 cm in length chosen, at random,
from a number of brands is unlikely to belong to the brand
in question. The purpose of statistics is to determine,
subject to various assumptions, exactly how likely it is that
our inference is correct (e.g., 90% sure; 99% sure).
“99% sure” is another way of saying that we expect our
inference to be proved wrong approximately only one time
in 100. Statistics is not, therefore, an inexact science, but the
science of being exact about the degree of uncertainty in a
statement which was based on numerical data. Statistics, of
course, cannot lie but those who use them can paint an
untrue picture by ignoring inconvenient measurements, by
making unjustified assumptions, and by omitting to tell the
WHOLE truth. By the end of this book, you will be better
equipped to test for yourself the validity of statistically
based statements.
Introduction ▪ xix
Now let us examine the meaning of the word Probability.

The probability of an event can be thought of as the ratio of
the number of occurrences of that event (i.e., successes) to the
total number of items of data, providing that the experiment
continues indefinitely. We often wish to determine the pro-
bability of some complex situation occurring where the inter-
action of a number of events is needed, each of which whose
probabilities are known. Probability theory involves mani-
pulating certain rules in order to determine the likelihood of
such outcomes.
Since, as we saw in an earlier paragraph, statistics involves
stating the probability of some numerical inference being
correct then, clearly, there is a link between statistics and
probability theory.
Since an estimate of probability is given by the ratio of the
number of occurrences (i.e., successes) of an event to the
number of items, it can be arrived at in two ways.
If it is established that, for UK males, 10% exceed 6 feet in
height then, from that information, the probability that a
person, chosen at random from a group of UK males,
exceeds 6 feet in height is indeed 10% (i.e., 0.1). This is an
empirical statement of probability being derived from
observed data.
If, on the other hand, we wish to know the probability of
drawing a heart from a deck of playing cards it can be
argued that since 25% of the cards are hearts then a card,
drawn at random, has a 0.25 probability of being a heart.
This is a probability statement based on prior knowledge of
xx ▪ Introduction
the population and NOT by experimentally drawing cards

repeatedly in order to determine the proportion of hearts.
We call this an à priori statement of probability.
In both cases, the concept of probability is derived from a
proportion of successes and is thus a dimensionless quantity
(that is to say it takes no units). A probability can take any
value between zero (impossibility) and unity (certainty).
Spreadsheets are an excellent and time-saving way of
carrying out statistical calculations. Those readers who are
not regular spreadsheet users would be wise to study
Appendix 1 carefully. It provides a “refresher” in manipul-
ating numbers quickly. Many of the exercises throughout this
book use the spreadsheet format and the reader is
encouraged to use one when solving them.
Now, read on!
Chapter 1
Why One Needs

Statistical Techniques
We are frequently faced with numerical information (known
as data) which, in its raw form, gives no clear picture of the
trends that it might contain. The following simple example
illustrates the point, as well as introducing two basic concepts.
The blood sugar levels in mmol/L (millimoles per litre),
recorded by a diabetic patient over a period of 10 days, were
7.5, 7.2, 8, 6, 5.6, 5, 7, 5.5, 5, 5.5
The average can easily be calculated as 6.2 although it may

not be obvious merely by glancing at the row of numbers.
Furthermore, the graph (Figure 1.1) shows that the trend
is, to some extent, downwards.
Thus, albeit at a very basic level, two important ideas
have emerged:
An average (we use the word MEAN in statis

tical work)
and
Providing a visual plot of the trend (a linear
GRAPH)
DOI: 10.4324/9781003220138-1 1
2 ▪ Statistical Techniques for Medical & Other Professionals
BLOOD SUGAR
10
6
mmol/L
0
Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7 Day 8 Day 9 Day 10
Figure 1.1 Simple graph of a variable against time
Both can be extremely helpful.

Notice the use of the word “linear” in the brackets. Linear
means that the distances separating the days along the x
axis and the distances separating the blood sugar levels,
ascending the y axis, are equal. This will not always be the
case in more advanced uses of the “graph” technique
(Chapter 5). We will return to that subject later.
Taking the concept of a MEAN further we might add,
providing the spread of values on either side is similar, that
we can make statements such as “I am 50% sure (in sta
tistics we use the word “confident”) that any value, taken
at random, is greater than 6.2.” That idea will be developed
in Chapter 4 to make more sophisticated inferences, such
as being 90% confident of exceeding some stated value.
In order to do that, we need to think about the variability of
the data. In other words, the spread (“distribution”) of values
between the two extreme values which were the two sugar
Why One Needs Statistical Techniques ▪ 3
levels of 5 and 8 mmol/L in the earlier example. Chapter 4 will

develop this idea further and later Chapters will provide
techniques for making useful comparisons between different
sets of data.
Variables and Attributes
At this point, it will be useful to distinguish between

Variables and Attributes.
Variables are a measure of an item of interest (e.g., size,
weight, time, blood sugar, systolic blood pressure) and can
take any value in a given range (e.g., feet, ounces, min
utes, mmol/L [blood sugar], mmHg [blood pressure].
Attributes, however, are binary and describe some state
that either applies or does not apply. In a deck of cards
any one card can be a heart (or not). A person can be
either alive or dead. One cannot be a bit dead, only alive
or dead. Thus, attributes are measured in numbers of items
having that attribute (or state) and have no units.
Spread of a Variable
The blood sugar readings illustrated in Figure 1.1 are dis

tributed (spread) around the mean value of 6.2 mmol/L.
The “tightness” or “looseness” of that spread is important
since it describes the consistency (or otherwise) of the
measurement in question. In the blood sugar example, a

consistent reading, shown by a tighter scatter, would be
desirable. Statistics will provide a way of describing how
consistent the readings are.
Correlation
Another problem, which statistics will help to unravel, is

the need to establish if some variable is related to another.
One example would be diastolic blood pressure and sys
tolic blood pressure. One does not precisely dictate the
value of the other, but it might be credible to assume that a
change in one might be followed by a similar change in
the other. Again, statistics will provide a more precise way
of quantifying the strength of the interaction. This parti
cular suggestion will be dealt with in Chapter 9 where we
will test the strength of association between the two.
Taking Samples
Much of statistical analysis consists of drawing conclusions

from a set of data or of comparing two or more sets of data.
The data in question is nearly always a sample drawn from
a wider (larger) population. It is therefore important to
think about whether the data that one has gathered is, in
deed, a representative sample of the population of interest.
Why One Needs Statistical Techniques ▪ 5
Random sampling is the ideal way of collecting data

from a population. To achieve this, each item in the po
pulation needs to have an equal chance of being chosen
for the sample. If, for example, the total annual population
of stroke sufferers is 115,000 (UK) and in conducting a
comparison between two alternative treatment regimes,
two sample groups of 50 patients were selected, there is
always the possibility that the sample did not represent the
population as a whole.
It is important to take an unbiased random sample from
which to draw conclusions. Thus, when we selected the
two samples of 50 patients from the total population of
115,000, there was the possibility that the researcher might
have selected 50% men and 50% women for the sample
even though the stroke population might contain say 67%
men and 33% women. The sample would not then accu
rately reflect some gender-related factor.
Gender is, of course, not the only relevant factor since
age, ethnicity, lifestyle, diet and occupation might all be
argued to be relevant. The problem with random sampling
is that it requires a complete knowledge of the population
before selecting the sample.
Very Large and Very Small Numbers
In many fields, data involves numbers in the hundreds of

thousands and greater and deals with factors such as one
in a hundred thousand and so on. “One in a hundred

thousand” is a cumbersome way of expressing data and for
those not familiar with the convention of expressing
numbers as positive and negative powers of ten, Chapter 9
provides a thorough grounding.
Chapter 2
Probability and Its

Rules
Empirical versus À Priori
An estimate of the probability of an event is given by the

ratio of the number of occurrences (i.e., successes) of that
event to the number of items of data, and this can be ar
rived at in two ways. Revisiting what was mentioned in the
introduction:
If it is established that, in the UK (population 67 Million),
there are 4.7 Million diabetics then, from that information,
the probability that a person, chosen at random, is diabetic
is 7% (i.e., 4.7/67). This is a probability statement based on
prior knowledge of the population and NOT by experi
mentally observing a sample. We call this an à priori
statement of probability.
If, on the other hand, we estimate the probability by
observing that, in a large medical practice covering 8,000
patients, there are 480 diabetics, then we might infer that
the probability of being diabetic is 6% (i.e., 480/8,000).
This is an empirical statement of probability being de
rived from sample data.
DOI: 10.4324/9781003220138-2 7
In both cases, the concept of probability is derived from a

proportion of successes and is therefore a dimensionless
quantity (that is to say it takes no units). A probability can take
any value between zero (impossibility) and unity (certainty).
Combining Probabilities
Assume that the probability of a person, selected at

random, exceeding 5’6” in height (Event A) is 0.8 and that
the probability of a person attaining the age of 65 (Event
B) is 0.7. Figure 2.1 represents this state of affairs by what
is known as a Venn diagram. The fact that the circles
overlap represents the fact that the two events are not
mutually exclusive – clearly it is possible to exceed a
height of 5’6” and, also, to attain the age of 65 years. The
area outside the circles represents all possibilities which
are neither event A nor event B.
Figure 2.1 Overlapping events

Probability and Its Rules ▪ 9
The shaded (overlapping) portion of the diagram re

presents the possibility that both event A and event B will
take place and the reader will doubtlessly recognise this as
the situation to which the Multiplication rule applies.
If Pa (the probability of event A occurring) is 0.8

If Pb (the probability of event B occurring) is 0.7
Then the probability that both event A and event B will

occur is
Pab = Pa x P b = 0 . 8 x 0 . 7 = 0 . 56
In other words, 0.8 proportion of the 0.7 proportion of the

total number of trials will result in both A and B together.
The rule can be extended to calculate the probability that
all of n events will occur together.
Pan = Pa x P b x P c … … … x Pn
The above equations assume independence between the

events Pa, Pb and so on. Consider the following case. The
probability of a new-born cot death is approximately one
in 8,000 births. Thus, the probability of two babies dying in
the same family might be inferred to be one in 8,000 x
8,000 = one in 64 Million. This assumes that if a child in a
family dies of infant death syndrome, then it is no more
and no less likely that a subsequent child will die in that
way. However, there are reasons why the two events
might not be independent as, for example, a genetic factor,

or an environmental factor unique to that family.
A different calculation involves the probability of either
A or B or both (in other words at least one) of the events
occurring. The answer, in this case, is
P (a or b or both ) = Pa + Pb (Pa x Pb )
( More easily remembered as SU M PR ODUCT)
Examining this in terms of ratios, 0.8 of events will turn out to

be A and, of these, some will also be B. Furthermore 0.7 of
events will turn out to be B and, of these, some will also be A.
Those events which are both A and B have, therefore, been
included twice and must be allowed for by deducting Pa x Pb,
namely the proportion which are indeed both A and B.
Another way of arriving at the same answer is to
argue that:
1−Pa is the probability of event A NOT occurring

1−Pb is the probability of event B NOT occurring
Thus, (1−Pa) × (1−Pb) is the probability that an event
will be neither A nor B.
Therefore 1 − (1−Pa) × (1−Pb) is the probability of
either A or B or both.
However: 1 − (1−Pa) × (1−Pb) = Pa + Pb – (Pa x Pb)
as above.
It follows that for n events the Probability of one or more

occurring is
1 (1 P a ) (l P b ) (1 Pc ) ……. (l P n).
Mutually Exclusive Events
Let us now examine the case of mutually exclusive events.

Consider a deck of cards and let event A be the drawing of
a heart and event B the drawing of a black card. This time
there is no overlap since a card cannot satisfy both out
comes. This is illustrated in Figure 2.2.
The previous theorem does not apply here since the
probability of both event A and event B taking place is
zero. It follows that the probability of drawing either a
heart or a black card (event A or event B) is
Figure 2.2 Mutually exclusive events

Pa + Pb
and for multiple events, the probability of observing either

event A, or event B, or event C, or event D, etc., is
Pa + Pb + P c + Pd etc
In the playing card example Pa = 0.25 Pb = 0.5 and

therefore the probability of drawing either a heart or a
black card is 0.25 + 0.5 = 0.75.
Exercise 1 Manipulating Probabilities

Assume that:
Having dark hair is 80% likely.

Having blue eyes is 25% likely.
Being bald is 5% likely.
Being taller than 1.8 m is 50% likely.
What is the probability of the following:
1. Having dark hair and blue eyes?

2. Having dark hair or blue eyes?
3. Having dark hair and being bald?
4. Having dark hair and blue eyes and being taller than
1.8 m?
5. Having dark hair or blue eyes or being taller than
1.8 m?
6. Having dark hair and blue eyes, or being bald?
7. Having neither blue eyes nor not being bald?
Conditional Probabilities
The following example illustrates a state of affairs in which

the probability of an event is conditional on the probability
of some other factor. Assume that a person, at random, has
a 1% probability of suffering from a particular form of
cancer (in other words 1% of the population are known to
suffer). Assume, as is often the case, that there is a test to
determine if the cancer is present but that the test is not
perfect. Like many tests, it may return a false positive de
spite the subject not suffering from the condition and it
might also return a false negative in that it fails to detect a
real case. The picture is summarised in the following table
with the assumption that there is a 10% chance of a false
result (be it positive or negative).
Patient Has Patient Has No

Cancer (1%) Cancer (99%)
POSITIVE TEST (Suggests 90% True 10% False
cancer is present) Positive Positive
NEGATIVE TEST (Suggests 10% False 90% True
cancer is not present) Negative Negative
Given a Positive Test
If one is tested at random (given the 1% chance of having

the disease), the probability of being a true positive is 90%
of 1% = 0.9%.
By a similar argument, the probability of a false positive

is 10% of (100% − 1%) = 10% of 99% = 9.9%.
Therefore, the probability of seeing any positive result is
0.9% + 9.9% = 10.8%
Since probability is the ratio of an event in question to
the number of trials then, we can argue that the probability
of having cancer (given that one receives a positive
result) is
(Probability of a true positive)/(Probability of any

positive)
= 0.9%/10.8% = 8.3%
Given a Negative Test
If one is tested at random (given the 99% chance of not

having the disease), the probability of being a true nega
tive is 90% of 99% = 89.1%.
By a similar argument, the probability of a false negative
is 10% of (100%−99%) = 10% of 1% = 0.1%.
Therefore, the probability of seeing any negative result
is 89.1% + 0.1% = 89.2%
Since probability is the ratio of an event in question to
the number of trials then, we can argue that the probability
of not having cancer (given that one receives a nega
tive result) is
(Probability of a true negative)/(Probability of any

negative)
= 89.1%/89.2% = 99.9%
Comparison of the Two Outcomes
In the case of testing positive the chances of a result, de

spite no illness, are swamped by the false positives of 99%
of the population, resulting in a small probability (8.3%) of
actually having the illness. On the other hand, the negative
result is a more reliable indicator due to the small con
tribution of false negatives from those suffering from the
condition. This all rests on the proportion of the popula
tion suffering from the condition. In the case of 50%, then
both positive and negative tests would yield the same
confidence of being correct.
The reader may care to populate the above examples
with alternative data relating to other conditions, the data
required for the study being:
■ Percentage of the population suffering from the

condition
■ Percentage of false-positive tests (despite no condi
tion existing)
■ Percentage of false-negative tests (despite the condi
tion existing)
In the above examples, for simplicity, the same values for

false positive and false negative results were used. In
practice, it is frequently the case that they are not the same
and are somewhat unbalanced.
The above equations are an example of Bayes
Theorem. In summary:
Probability of Cancer (given a positive result )

= Probability of Cancer x
Probability of a positive result if Cancer is present
Probability of any positive result
The equivalent equation applies to the negative result.

Appendix 1 provides a spreadsheet aid to calculating
the results. It can be used to evaluate the data from
different tests.
Chapter 3
Dealing with Variables

In Chapter 1, the distinction between Variables and
Attributes was explained. In this chapter, we shall in
troduce ways of expressing numerical data as values of
some variable. Following that we shall address how to
draw conclusions from the data.
A variable is usually expressed in terms of some continuous
measurement such as feet, mmol/L [blood sugar content], etc.
That type of continuous variable can take any value in a range
and the number of significant figures is limited only by the
accuracy available to the measuring technique involved.
Metrics
There are, on the other hand, semi-quantified metrics, which

can also be treated as variables. Sometimes, we cannot
measure some specific feature of interest but have to rely on
subjective judgement. An example would be asking a patient
to assess perceived pain level using a score of 1 to 5 (1 being
minimal and 5 being extreme and unbearable). This type of
metric has no units but, nevertheless, we can treat it much
like a variable. Figure 3.1 shows an example.
DOI: 10.4324/9781003220138-3 17
Figure 3.1 Semi-quantified metric assessing pain
Figure 3.2 Normal (Gaussian) curve
In this chapter, we will confine ourselves to the specific

distribution (spread) of a measured value which is called
the NORMAL (or Gaussian) distribution. This applies
where the values are equally spread on either side of the
average (centre) value and form a symmetrical bell-shaped
curve the shape of which is illustrated in Figure 3.2.
The two most important “measures” which are relevant
to such a collection of values are
A measure of “central tendency” or “average”

called the
MEAN
Dealing with Variables ▪ 19
A measure of “scatter” or “spread” called the

STANDARD DEVIATION
The Mean
The Mean, or to be more specific, the Arithmetic Mean, is the

sum of all the values divided by the total number of individual
values. This is more concisely expressed as a formula:
Xi
X̄ =
N
The X with a bar on top signifies the Mean of the X values.

Xi signifies each of the N values of X. The Greek letter
sigma ∑ signifies adding together all the Xi values.
The Standard Deviation (and Variance)
This involves taking the difference between each value of

Xi and the Mean and then formulating a parameter to de
scribe the degree of scatter. That is the standard deviation
and the formula is
(X̄ Xi)2
N
Without the square root it is called the variance, in other
words:
(X̄ Xi)2
N
Grouped Data
Sometimes, in a set of data, we have more than one item

with the same value. In order to record these more suc
cinctly, each value is recorded once with a separate entry
to indicate how many such items. The following Exercise 2
will handle the data in that form.
A Range for Each Value
Where the data is expressed as an integer, then any number

in a range either side of the integer will take that integer
value. Thus, in the following Exercise 2, the number 7 de
notes being between 6.5 and 7.5 years of age and so on.
Making Inferences from the Mean and

Standard Deviation
We have seen earlier that, where there is a symmetrical

distribution, it is meaningful to state that the probability
(chance) of exceeding the mean value is 50%. This much is
fairly obvious. However, we are now in a position to make
Exercise 2 Obtaining the Mean and Standard

Deviation
Calculate the mean and standard deviation for the fol
lowing set of data. Make use of the information in
Appendix 1 to do this using a spreadsheet.
Age Number in Group
6 1
7 2
8 4
9 4
10 2
11 1
a far wider range of inferences. Figure 3.3 shows a normal

curve and introduces the two ideas of Inference and
Confidence.
Figure 3.3 Normal curve and data

Inference means that statements based on the mathe

matics of that curve will hold good for the data in question,
given that the data set is seen to be a good fit to the Normal
distribution curve. In Chapter 8, we will look at a method
of determining the “goodness of fit”.
Confidence is the term used to refer to the probability
(or chance) that a particular inference is true.
In the earlier statement about exceeding the mean then,
for Figure 3.3, a statement might be “50% Confident that a
person is more than 6 feet in height”. This is borne out by
the observation that 50% of the area under the normal
curve lies to the right of the mean.
By a similar use of the areas under the curve, it is
possible to state the likelihood (confidence) of being less
than, greater than, or falling between, any two values.
Appendix 3 provides a table of the normal distribution.
Look at Appendix 3 and, by taking differences, note that:
■ The range + or – 1 standard deviation occupies

68% of the curve
■ The range + or – 2 standard deviations occupies
95.5% of the curve
■ The range + or – 3 standard deviations occupies
99.8% of the curve
Therefore, in respect of Figure 3.3, it is possible to say that

we are
■ 68% confident that a person’s height is between 5’ 11”

and 6’ 1”.
■ 95.5% confident that a person’s height is between 5’
10” and 6’ 2”.
■ 99.8% confident that a person’s height is between 5’ 9”
and 6’ 3”.
Notice that these statements reflect a range and are thus

referred to as double-sided inferences.
We could, however, say that one is:
■ 99.9% confident that a person’s height is greater than

5’ 9”.
This latter statement would be referred to as a single-sided

inference.
The Coefficient of Variation
We are now equipped to make use of what is known as

the coefficient of variation (COV). It is simply the ratio of
the standard deviation to the mean. Expressed as a per
centage it provides a “feel” for the degree of scatter and
hence the consistency (or otherwise) of the set of values.
A COV of 5% would mean that 68% of the values fall
between + or – 5% of the mean. This would indicate a far
tighter (consistent) distribution than a COV of, say, 10%.
Figure 3.4 shows graphs relating to a patient’s blood

sugar over a nine-day period. One shows the trend in
mean sugar level and the other the trend in COV. As can
Figure 3.4 Graphs of Mean and COV blood sugar

be seen, not only has the sugar level tended to fall but so
also has the COV, indicating an improvement in con
sistency. The latter is as important as the trend in the Mean.
Exercise 3 Mean and Coefficient of Variation

The following mmol/L blood sugar measurements were
recorded by a patient. Calculate the mean and COV and
comment.
13- 14- 15- 16- 17- 18- 19-

Nov Nov Nov Nov Nov Nov Nov
6.90 6.87 6.83 6.90 6.33 6.63 6.67
Other Measures of Central Tendency

(Types of Average)
Medians
The median is that value on either side of which 50% of the

values lie. Thus,
1,2,3,7,8,10,11 yields a Median of 7 (note that the

mean is smaller, 6)
This is because the distribution is skewed to
the left.
Figure 3.5 Skewed distributions from left to right: The first ex
ample, an unbiased (symmetrical) distribution, and the second
example
1,2,4,8,10 yields a Median of 4 (note that the

mean is greater, 5)
This is because the distribution is skewed to the right.
Figure 3.5 illustrates, from left to right:
■ The first example

■ An unbiased (symmetrical) distribution
■ The second example
For a normal distribution:
The Mean = The Median
However, since the median provides the value which we

are 50% confident to exceed, it is useful for expressing
expectancy of survival.
Exercise 4 Median Life Expectancy

Following a specific diagnosis, the following table shows
[to the nearest year] how long each of 50 patients survived.
Calculate the median life and the likelihood of survival to
each year.
No of Patients Years of Survival

2 2
6 3
8 4
11 5
13 6
9 7
1 8
Note that each survival age signifies a range of plus or

minus 0.5 year on either side of the value stated.
Geometric Mean
The geometric mean of two numbers, X and Y, is calcu

lated as the Square Root of X times Y.
Geometric means are more meaningful than arithmetic
means in two areas:
a. Geometric Growth
If the growth of some variable is geometric rather than
linear (e.g., population) where the arithmetic mean
would be misleading. Thus, if the population of
Noddytown is 10,000 in 2010, and 20,000 in 2020, then
the likely number in 2015 would be better given as:
Square Root of 10,000 × 20,000 = 14,142.

Rather than 15,000 (arithmetic mean)
In Chapter 5, this will be illustrated graphically.
b. A wide range
Assessing the most likely value when we only know a

wide range (e.g., 1 to 100).
Again, the square root of 1 × 100 = 10.
This is arguably more meaningful than taking the ar
ithmetic mean, of approximately 50, which is 50 times the
smaller value but only half of the larger value.
Chapter 4
Comparing Variables
In Chapter 3, we acquired some familiarity with the normal
(Gaussian) distribution and its mean and standard deviation.
The next step is to think about comparing two dis
tributions (of the same variable) with a view to deciding if
they are significantly different or whether they might both
represent the same population.
Figure 4.1 illustrates the idea graphically. The two
questions are
a. Are the standard deviations significantly different?
b. Are the means significantly different?
Comparing the Standard Deviations
A factor known as F (from Fisher’s F distribution) is cal

culated as the ratio of the larger to the smaller of the two
variances. Remembering that the variance is the square of
the standard deviation, then:
StdDev12
F =
S td De v22
Appendix 4 provides tables of the F distribution. In order
to use them, it is necessary to make use of a number
DOI: 10.4324/9781003220138-4 29
(a) (b)
Significantly different?? Significantly different??
Figure 4.1 Comparing distributions: a) Are the standard deviations

significantly different? b) Are the means significantly different?
known as the degrees of freedom. It is simply n‒1, for each

of the variances, where n is the number of items in each of
the two distributions being considered.
Assume that there are two distributions that we wish to
compare, where F = 3.5 and the two sample sizes (n) are
both 10. Looking at the 2.5% table (Appendix 4b) it can be
seen that the nearest entry to 9 degrees of freedom (n‒1) is
given by column10/row10 which gives a value of 3.72. The
table tells us that there is, therefore, an approximately 2.5%
chance that the two distributions have been drawn from a
common population. In other words, it is 97.5% likely that
they are different.
Imagine, for a moment, that the value of F were to be
5.0 then, looking at the 1% table (Appendix 4a) it can
be seen that the nearest entry to 9 degrees of freedom
(n‒1) is 4.85. The table tells us that there is, thus, an
Comparing Variables ▪ 31
approximately 1% chance that the distributions are from a

common population. In other words, it is 99% likely that
they are significantly different.
Notice that larger values of F indicate a greater significance
of difference.
Comparing the Means
A factor “t” (from the Students t distribution) is calculated

from the difference of the means divided by the square
root of [StdDev12/n1 + StdDev22/n2]. Hence,
[M ean1–Mean2]
t=
SQR T [StdDev12 / n1 + StdDev22 /n2]
The degrees of freedom are given by n1 + n2 – 2

Looking at the t table (Appendix 5) and, assuming that
t = 0.7 (with 10 degrees of freedom), we can see that there is
only a 50% likelihood that the distributions are significantly
different. As the value of t increases, then so does the like
lihood of a significant difference.
We can now proceed to the next Exercise making
use of the F tables in Appendix 4 and the t Tables in
Appendix 5.
Exercise 5 Comparing Two Distributions

Two treatment regimes, for Parkinson’s disease, yielded
the following response times in a reaction test.
For each of 6 patients before treatment:
61, 32, 37, 43, 29, 50 seconds
For each of those 6 patients after treatment:

45, 30, 32, 36, 27, 43 seconds
Determine if the variances and means are significantly

different and form an opinion about the effectiveness, or
otherwise, of the treatment.
Double- and Single-Sided Inferences
In Chapter 3 (Figure 3.3 and subsequent paragraphs), we

saw how single-sided statements (such as “greater than”,
“less than”) and doubled-sided statements (“between”) can
be made by choosing the appropriate areas of the normal
distribution curve (Appendix 3). In much the same way,
we can do this with the “t” distribution.
The double-sided use of “t” (which are the tables in
Appendix 5) provide inferences that the difference is (or is
not) significant in respect of “one mean greater” or “the other
mean greater”. In most cases, this is the interpretation we
require. The question of whether the difference is desirable
or undesirable, in the case of the treatments/options being
compared, is a matter for the analyst to interpret.
Chapter 5
Presenting Data and

Establishing Trends
Visual Presentation of Data
Histograms
Statistics is concerned with analysing numerical data, so

the way in which the data are presented is of great im
portance. We are frequently told “A pictures paints a
thousand words”. Consider Table 5.1 which gives 100
values of a hypothetical parameter involved in some as
pect of haematology. The 100 values are presented in 10
rows of 10 and in the order in which they were observed.
The appearance of the above table of values provides
no picture of the way in which the values themselves are
distributed. It is not easy to see which value occurs most
frequently, or which value indicates the centre point of the
distribution or, for that matter, whether the values are
distributed symmetrically about the centre. It is very likely
that we would want some idea of the degree of scatter
(deviation) of the values and this “picture” is certainly not
evident from the table.
DOI: 10.4324/9781003220138-5 33
34 ▪
Table 5.1 An Example of Some Data

140 155 335 211 211 198 323 294 321 235
251 271 181 227 361 306 221 252 311 165
255 255 273 293 181 259 258 236 237 211
191 294 256 273 272 220 274 256 220 301
237 244 275 258 276 220 192 192 237 241
260 221 192 292 189 301 279 281 241 231
232 220 261 281 282 242 262 225 283 264
226 284 243 265 227 208 311 245 286 266
247 246 267 289 247 227 228 268 322 248
172 335 208 209 291 269 249 249 269 192
Statistical Techniques for Medical & Other Professionals
Presenting Data and Establishing Trends ▪ 35
Table 5.2 Grouping and Ranking the Data

Class Boundaries Frequency
130–149 1
150–169 2
170–189 4
190–209 9
210–229 15
230–249 19
250–269 19
270–289 15
290–309 8
310–329 5
330–349 2
350–369 1
100
Table 5.2, however, shows the same values arranged in

a more orderly fashion. Three features distinguish the two
Tables 5.1 and 5.2. In the second, table the range of values
is divided into class boundaries such that all values falling
within a particular class are treated as similar. Secondly,
each class is indicated once only and the number of values
falling within each class is indicated in the frequency
column. Thirdly, the classes are ranked from lowest to
highest.
Figure 5.1 The data as a histogram
The central tendency of the values, their range and the

way in which they are clustered around the central value are
somewhat clearer as a result of their grouping in Table 5.2.
Better still, in Figure 5.1, this information is presented in
the form of a bar chart (HISTOGRAM) in which each
class is represented by a bar whose height is proportional
to the number of values within that class.
The question arises as to how many classes should be
chosen into which to allocate the values. If the number of
classes is too high, relative to the number of values in the
data set, then there will be too few values in each class and
the histogram will appear flat as seen in Figure 5.2.
If, on the other hand, there are too few classes, then the
histogram will look like Figure 5.3.
Figure 5.2 Histogram with too many classes
Figure 5.3 Histogram with too few classes

Figure 5.4 Data showing the superimposed normal distribution
The shape of the histogram in Figure 5.1 suggests a

normal distribution as illustrated in Figure 5.4. Chapters
3 and 4 dealt with establishing the mean and standard
deviation parameters for the curve. Chapter 8 deals with
establishing if the normal distribution curve is (or is not) a
good fit to the data set.
Graphs
The previous section has dealt with a single variable.

Frequently we are presented with data concerning two,
possibly linked, variables. The following Table 5.3 shows
24 blood pressure readings in mmHg. Again, the table
alone tells us very little.
Table 5.3 24 Blood Pressure Readings

Systolic Diastolic Systolic Diastolic
148 73 144 82
132 61 167 77
139 73 134 64
155 66 140 64
136 66 129 66
133 64 150 63
162 73 115 62
142 71 118 64
128 58 134 67
136 62 132 61
146 54 122 61
143 54 120 56
Figure 5.5 shows how, by representing each reading as

a Y and an X value on a diagram, a visual picture of their
relationship emerges.
The danger comes when we try to express the re
lationship more precisely by drawing a straight line (i.e., a
regression line) which appears to best fit the data.
Figure 5.6 shows how several possible regression lines
might each claim to serve that purpose. The technique,
although useful in terms of a general picture has, never
theless, the potential to be misleading.
Figure 5.5 Scatter diagram
Figure 5.6 Scatter diagram with regression lines
In Chapter 8, we shall deal with correlation and re

gression where a more precise method of assigning the
line will be explained. However, this can then to lead to
the bad practice of showing the line without the scatter
points, thus giving a false impression of the strength of
association of the two variables.
Some Pitfalls
Supression
The following graph (Figure 5.7) shows the coefficient of

variation of a patient’s blood sugar over four months,
giving the impression of a very significant trend of im
provement. Note, however, that the vertical axis only
displays a range from 12% to 21%.
Figure 5.8 shows the true picture, which is not as
spectacular as might have been inferred, when supressing
the zero.
Figure 5.7 Trend in COV

Figure 5.8 Trend with zero no longer supressed
Extrapolation
Figure 5.9 illustrates the danger of assuming that a trend will

continue unaltered. The data in Figures 5.7 and 5.8 only
provide four consecutive measurements. There is no
Figure 5.9 Example of extrapolation

guarantee that the trend will continue, in that the circum

stances pertaining to the days beyond Day 4 may change,
such as to alter the trend. Nevertheless, extrapolation of
graphs is common. This practice is sometimes justified but
needs to be argued (i.e., justified) for each and every case.
Logarithmic Scales
The difference between a linear and a logarithmic scale is

illustrated by Figures 5.10 and 5.11. Figure 5.10 shows the
growth in population of some hypothetical country. The
graph illustrates, not unrealiastically, the accelerating
nature of the growth.
Figure 5.10 Population growth (linear scale)
Figure 5.11 Population growth (logarithmic scale)

Figure 5.11, however, plots the same data on a loga

rithmic scale. That is to say equal distances, ascending the
y axis, represent equal MULTIPLES rather than equal
INCREMENTS. The picture is quite different. It suggests a
more or less constant rate of growth.
Both representations are valid in their own way de
pending on whether the intention is to illustrate the ab
solute number (Figure 5.10) or the rate of growth
(Figure 5.11).
It reminds us to be careful when looking at graphs to
check what linearity applies to the axes since, improperly
presented, a picture could be misleading.
Figures 5.12 and 5.13 provide a further example where,
this time, the graph shows a decreasing quantity. The
number in question is a moving average which will
be dealt with in the next section. The point here is that the
DAILY DEATHS 7 DAY MOVING AVERAGE

1,400.0
1,200.0
1,000.0
800.0
600.0
400.0
200.0
0.0
01/02/2021
06/02/2021
11/02/2021
16/02/2021
21/02/2021
26/02/2021
03/03/2021
08/03/2021
13/03/2021
18/03/2021
23/03/2021
28/03/2021
02/04/2021
07/04/2021
12/04/2021
17/04/2021
22/04/2021
27/04/2021
02/05/2021
07/05/2021
12/05/2021
17/05/2021
Figure 5.12 Moving average of deaths on a linear basis

DAILY DEATHS 7 DAY MOVING AVERAGE

10,000.0
1,000.0
100.0
10.0
1.0
01/02/2021
06/02/2021
11/02/2021
16/02/2021
21/02/2021
26/02/2021
03/03/2021
08/03/2021
13/03/2021
18/03/2021
23/03/2021
28/03/2021
02/04/2021
07/04/2021
12/04/2021
17/04/2021
22/04/2021
27/04/2021
02/05/2021
07/05/2021
12/05/2021
17/05/2021
Figure 5.13 Moving average of deaths on a logarithmic basis
decrease in average UK COVID death rate, from mid-

January onwards, appears to become less and less when
expressed on a linear axis (Figure 5.12). The logarithmic
picture (Figure 5.13), however, shows a fairly constant
relative drop as time goes by.
Moving Averages
The following graph (Figure 5.14) shows the variation of

daily UK fatalities during the 2020 COVID pandemic.
Daily variations (during each week) and the effect of
weekends give rise to a ragged picture. If the mean
(average number of fatalities) is calculated for a moving
window of 7 days (Figure 5.15), the picture becomes
clearer.
Figure 5.14 Daily data (unsmoothed)
Figure 5.15 Daily data (smoothed)
In Figure 5.15, a 7-day window was used due to the

irregularity of the numbers published at weekends. The
“weekend distortion” effect was, thus, smoothed out.
In the following example (Figure 5.16), there are 16
consecutive values of a patient’s estimation of pain.
The graph of raw data shows no obvious trend.
Figure 5.16 Charts of pain estimation
However, by taking a 4-day moving average a trend

emerges. The first value of the moving average, in this
case, is constructed by taking the average of the first 4
values (6, 6, 4, 8). The next removes the “6” value and
includes the “5”. The window thus moves to the right
selecting 4 values at a time. The choice of a 4-day
window is arbitrary and should be made having regard
to the circumstances.
Control Charts
These are graphs of a chosen variable with warning and

action lines to stimulate responses to undesirable var
iations. Figure 5.17 shows a simplified blood pressure
SYSTOLIC
250
200
150
100
50
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
ACTION WARNING
LINE LINE
Figure 5.17 Blood pressure control chart in mmHg
chart with action and warning lines. The choice of va

lues for each line may be a matter of judgement or, if
data is available, from previous outcomes involving
those levels.
Chapter 6
Dealing with Attributes

Up to now we have been dealing with Variables where the
data is expressed as values of some item of interest (e.g.,
weight, size, blood sugar values). As we saw, at the beginning
of the book, Attributes involve integers whereby we count the
number of items to which the specific Attribute applies.
Binomial
Think about the following experiment with a deck of

52 playing cards. A card is removed at random, its suit
noted, and then replaced. A second card is then removed,
and its suit also noted. The possible outcomes of this
experiment, in respect of hearts, are as follows:
H&H: H&notH; notH&H; notH&notH
where ‘H’ signifies a heart and ‘notH’ signifies a card other

than a heart.
The probability of drawing a heart from a normal deck
of cards is 0.25 (P). The Probability of drawing a card of
some other suit is therefore:
DOI: 10.4324/9781003220138-6 49
1 – 0.25 = 0.75 (Q).
Note that P + Q = 1
From the multiplication rule in Chapter 2
■ The probability of drawing 2 hearts is P × P = P2 = 0.0625

■ The probability of drawing a heart as the first card and
some other suit as the second card is P × Q = 0.1875
■ The probability that the first card is not a heart and
that the second is a heart is Q × P = 0.1875
■ The probability that both are not hearts is Q × Q =
Q2 = 0.5625.
Expressing this in general we can say:
■ Probability of 2 hearts P2
■ Probability of 1 heart 2PQ
■ Probability of 0 hearts Q2
In the same way, for 3 cards the result will be:
■ Probability of 3 hearts P3
■ Probability of 2 hearts 3P2Q
■ Probability of 1 heart 3PQ2
■ Probability of 0 hearts Q3
The reader will recognise the above probabilities as the

terms of the expressions (P + Q)2 and (P + Q)3. This leads
Dealing with Attributes ▪ 51
to the general statement that if P is the probability of some

random event, and if Q is 1 ‒ P, then the probabilities of
0, 1, 2, 3, etc., outcomes of that event, in n trials, are given
by the terms of the expansion:
(P + Q)n = Pn , nP(n 1) Q, n(n 1)P(n 2) Q2 …… ……. etc.

2!
This is known as the Binomial expansion. It has been in-

cluded for general background (depth of knowledge) and
interest. In most cases, we will be able to move on to the
next section which provides a simpler approach.
Poisson
In the case of attributes (which is what we are dealing with in

this Chapter) items are classified as either having the attribute
or not. So for a sample of n, drawn from a population con-
taining the P proportion, probabilities of seeing 0, 1, 2, 3, etc.,
are given by the terms of the above Binomial expansion.
Usually, P is a very small number (e.g., 0.05 which is 5%) and
n is a reasonably large number say >5. In such cases an ap-
proximation known as the POISSON DISTRIBUTION applies.
The great advantage of the Poisson Distribution is that,
whereas the Binomial is specified by 2 quantities (P and n),
the Poisson requires only the single number given by the
product of nP. We call nP the “expectation” since it is the
expected average number of items, in the sample, which have
the specified attribute. This does not infer that every sample
will contain the quantity given by exactly n × P, indeed some
will contain 0, others 1, others 2 and so on. If the expectation
is given the symbol m such that m = nP then the terms of the
Poisson Distribution are:
e−m, me−m, m mn e m
m2 e
, n!
2!
For each of:
0 1 2 n items
where e is a number (often used in mathematics) whose

value is approximately 2.7. It is not necessary to remember
the value of e, or the terms of the Poisson expansion be-
cause, as is the case with the Normal, F and t Distributions,
tables are available. In the case of the Poisson Distribution
these tables are often expressed in the form of a set of curves
which give the probability of n or less items, having the
attribute in question, plotted against m for different values of
n. A little caution is necessary since some tables and curves
give the probabilities of n or more occurrences whilst others
give those of n or less and others the probability of exactly n.
Appendix 6 gives curves of the probabilities of n or less
occurrences over a range of values of m from 0.1 to 20 and
for values of n from 0 to 20.
Dealing with Attributes ▪ 53
In the following example, the annual probability of fatal

influenza is 1 in 4,000 which is better expressed as 2.5 10−4
(Chapter 9 deals with the practice of expressing numbers
using exponents).
Thus, in a town of 10,000 inhabitants, the anticipated
number of fatal cases is m = 2.5. From the curves
(Appendix 6) the following probabilities can be obtained:
0 deaths is 8%
1 or less deaths is 29%
Exercise 6 Attributes Using Poisson Curves

A practice cares for 2,000 potential patients.
It is known that the incidence of stroke, in the UK, is
approximately 115,000 pa (population 67 million).
What is the probability that the practice will need to treat?:
0 cases?
1 case?
2 cases?
3 cases?
4 cases?
5 cases?
Chapter 7
Testing for
Significance
(Attributes)
We are frequently confronted with data relating to some
attribute where measurements have been taken under two
different regimes or circumstances. The following tech-
nique applies to the numbers of occurrences of an attri-
bute and is used to compare the difference between
outcomes in terms of whether that difference is significant
or is merely due to chance.
It is best explained by means of the following example.
Assume that we wish to test if the observed frequency of
seeing each face of a die is random or whether that par-
ticular die is biased.
We calculate the value of what is known as χ2 called
Chi-squared (χ is the Greek letter Chi pronounced Kai, as
in the word kite). Some books refer to it as Chi- square. It
is calculated by comparing the observed frequencies (O)
with the frequencies which could be anticipated if the
outcome were assumed to be unbiased (A). The formula is
DOI: 10.4324/9781003220138-7 55
2 (O A)2
=
A
Compare the observed values from throwing the die 60

times with what might be assumed if it were to be perfectly
unbiased. The first step is to list the observed frequency of
“seeing” each face. Thus, column 1 of Table 7.1 denotes
the face and column 2 the number of observations. In
column 3, we have entered the anticipated number of
occurrences if the die were perfectly balanced, in which
case the 60 throws would lead to 10 occasions per face.
Column 4 records the difference between the observed
(O) and the anticipated (A). Column 5 squares the differ-
ence. Column 6 divides this square by the anticipated
Table 7.1 Outcome from Sixty Throws of a Die

Observed Anticipated
Face Frequency Frequency
0 A O-A (O-A)2 (O-A)2/A
1 10 10 0 0 0
2 8 10 −2 4 0.4
3 6 10 −4 16 1.6
4 11 10 1 1 0.1
5 13 10 3 9 0.9
6 12 10 2 4 0.4
3.4
Testing for Significance (Attributes) ▪ 57
result. The sum of the column 6 values provides the value

of χ2.
The χ2 value of 3.4 is compared (using the Appendix 7
table) with what might be anticipated if the result were
purely random. Since there were 6 faces there are 5 in-
dependent values and this is called the “degrees of
freedom” namely n-1.
Row 5 of Appendix 7 tells us that 3.4 is between 70%
and 60% likelihood. Therefore, we infer that it is ap-
proximately 65% likely that the difference is due purely to
chance and we might reasonably infer that the die is not
biased.
The above example compared a set of outcomes with
an assumed hypothesis. In Exercise 7, we will use the
technique to compare to sets of outcomes resulting from
different treatments
58 ▪
Exercise 7 Chi- Squared

Comment on the effectiveness of the following two treatment regimes for diabetes.
TREATMENT A WITH INSULIN ONLY
BASE DATA TREATMENT A ANTICIPATED
IMPROVEMENT IMPROVEMENT IF NO EFFECT
> 10% REDUCTION > 10% REDUCTION TREATMENT A
W/O TREATMENT WITH TREATMENT
1 MONTH 4 2 1.655172
1–2 MONTHS 15 6 6.206897
2–3 MONTHS 6 2 2.482759
>3 MONTHS 4 2 1.655172
29 12 12
Statistical Techniques for Medical & Other Professionals
TREATMENT B WITH INSULIN & ADDITIONAL TREATMENT
BASE DATA TREATMENT B ANTICIPATED
IMPROVEMENT IMPROVEMENT IF NO EFFECT
> 10% REDUCTION > 10% REDUCTION TREATMENT B
W/O TREATMENT WITH TREATMENT
1 MONTH 4 0 2.344828
1–2 MONTHS 15 8 8.793103
2–3 MONTHS 6 5 3.517241
>3 MONTHS 4 4 2.344828
29 17 17
Testing for Significance (Attributes)
▪ 59
Chapter 8
Correlation and
Regression
Relating Two Variables
Having dealt with attributes, we now return to looking at

variables and would like a technique that provides a metric
for describing the strength of association between two
variables.
In Figure 8.1, the scatter chart shows a number of blood
pressure readings that record the systolic and diastolic
blood pressures. The point of interest is whether the one is
a reliable predictor of the other. The chart shows some
(albeit loose) correlation between the two and the ques
tion is “what is the strength of that association”. The lines
drawn through the data show that the association is not
exact. In fact, we could potentially draw several different
lines which look equally convincing.
The technique is to fit a line to the data which best
models the fit. The metric that is used to describe the
“goodness of fit” to the data values is called the correlation
coefficient and it is given the symbol “r”. The formula is
DOI: 10.4324/9781003220138-8 61
Figure 8.1 Two possible regression lines (visually)
not simple and is dealt with in References 1 and 2 (see

Appendix 9).
r = 1 signifies 100% correlation as, for example, be

tween water pressure and water depth
r = 0 signifies no correlation at all (in other words
randomness of association between the two variables)
In practice, we would be looking for r2 > 0.95 for a robust

association (r2 is referred to as the coefficient of determi
nation). For those readers who would like to take the
subject a little further the formula is given below:
(xi xmean)(yi ymean)/SquareRoot[ (xi xmean)2 (yi ymean)2)]
The Greek sigma, ∑, signifies sum all of the values.

Figure 8.2 is a spreadsheet set to calculate “r”.
Correlation and Regression ▪ 63
Figure 8.2 Regression spreadsheet
As can be seen in Figure 8.2, the r value of barely 0.5

signifies a very poor correlation between the systolic and
diastolic readings in respect of that particular set of data.
The regression equation is a straight-line expression of
the form:
Y= mX+ c
It can be constructed from the data and this is dealt with

in References 1 and 2 (see Appendix 9). In addition,
Appendix 1 explains the spreadsheet function for
obtaining this.
Exercise 8 Correlation
Construct the spreadsheet shown in Figure 8.2 using
the formulae indicated. Cell D32 ( Figure 8.2) contains
the equation for r provided above the figure. Enter the
following data and determine the coefficient, r.
Systolic Diastolic
147 73
156 78
160 73
189 80
173 68
157 70
139 80
157 79
165 60
156 82
175 78
158 70
161 80
180 74
145 78
162 69
(Continued)
Systolic Diastolic
147 70
168 79
131 68
138 75
126 58
147 73
145 70
145 69
False Correlation
Beware of false correlations where a large value of r is

obtained but, in fact, the two variables are only linked by
virtue of some common factor.
For any two correlated events, A and B, their possible
relationships include the following:
■ A causes B (direct causation);

■ B causes A (reverse causation);
■ A and B are both caused by C;
■ A causes B and B causes A;
■ There is no connection between A and B; the corre
lation is a coincidence.
Thus, there can be no conclusion made regarding the

existence or the direction of a cause-and-effect re
lationship only from the fact that A and B appear to
be correlated. Determining whether there is an actual
cause-and-effect relationship requires further investiga
tion, even when the relationship between A and B is
statistically significant.
Establishing a Normal Distribution
We may have some data giving us values for a specific

variable and wish to establish if the distribution is, in fact,
Normal. The following example is a list of 25 data values.
The procedure is to:
■ Rank the values from lowest to highest.

■ Establish the median rank for each by what is known
as Benard’s approximation [(Rank-0.3)/(N + 0.4)].
■ Use the Excel ‘NORMSINV’ function to express each in
terms of its equivalent value as a number of standard
deviations from the mean.
■ Use the correlation tool (Exercise 8) to examine the
goodness of fit.
The following spreadsheet shows how the NORMSINV is

obtained for each of the 25 ranked items (Figure 8.3).
No of STDEV
Median away from Ranked
Rank Mean Data
Rank "NORMSINV"
1 0.0275591 -1.917944016 1
2 0.0669291 -1.499059227 2
3 0.1062992 -1.246452206 2
4 0.1456693 -1.055189681 3
5 0.1850394 -0.896325881 3
6 0.2244094 -0.757385572 3
7 0.2637795 -0.631736528 4
8 0.3031496 -0.515363243 4
9 0.3425197 -0.405596134 4
10 0.3818898 -0.300521332 4
11 0.4212598 -0.198671544 5
12 0.4606299 -0.098846884 5
13 0.5 0 5
14 0.5393701 0.098846884 5
15 0.5787402 0.198671544 5
16 0.6181102 0.300521332 6
17 0.6574803 0.405596134 6
18 0.6968504 0.515363243 6
19 0.7362205 0.631736528 6
20 0.7755906 0.757385572 7
21 0.8149606 0.896325881 7
22 0.8543307 1.055189681 7
23 0.8937008 1.246452206 8
24 0.9330709 1.499059227 8
25 0.9724409 1.917944016 9
Mean 5
Std Dev 2.041241
Figure 8.3 Establishing NORMSINV
The NORMSINV values are transferred (use paste special,

values) to the following spreadsheet and correlated against
the 25 data values in question. In the example, a correlation
Figure 8.4 Establishing the regression coefficient
coefficient, r, of 99% is obtained. Thus, r2 is 98% that in

dicates an excellent fit to the normal distribution. Inferences
may therefore be made on the assumption that the data is
normally distributed (Figure 8.4).
Chapter 9
Handling Numbers
(Large and Small)
If the reader is not already familiar with manipulating
numbers expressed in the “POWERS OF TEN” format, then
it is important to study this chapter carefully and to prac-
tice Exercise 9.
Many of the numbers involved in this subject are either
very large or very small (in other words, a number multi-
plied or divided by a large number, like a million).
If we use expressions such as “1 in 1,000,000” or “1 in
100,000” it becomes very cumbersome. It is, therefore,
important to become familiar with the concept of
“POWERS OF TEN.”
Big Numbers
The “power” of ten means the number of tens multiplied

together so:
DOI: 10.4324/9781003220138-9 69
10 times a year means 10 occurrences each year. There is only 101

one “ten” so we write this as 10 to the power of 1. Maybe
somewhat obvious but it sets the pattern for what follows
100 times a year is 10 × 10 each year so let us call it 10 squared 102
or “10 to the power of 2”, or even more slickly, 10 to the 2
In the same way, 1,000 each year becomes 10 x 10 x 10 or “10 to 103
the 3” per year
In general, therefore, 10 multiplied by 10, N times is

expressed as 10N. You saw that 10 is written as 101. We
usually omit the “1,” but it has a purpose as was seen
above. In the same way, the number one can be written as
100 (i.e., 10 to the power of zero). We sometimes need to
calculate the square root in which case the exponent
becomes 0.5 (i.e., 1/2). Thus, the square root of 10 is 100.5.
Small Numbers
Now we shall look at numbers much smaller than one. This

time it becomes necessary to express them in a way which tells
us how many times to DIVIDE (rather than multiply) by 10.
1 in 10 years is statistically “one tenth” of an incident per year or 10−1

“one divided by 10” so we call it 10−1. In this context the minus
signifies “divide”.
In the same way 1 in 100 becomes one in “10 to the 2”, i.e., 10−2 10−2
Similarly, “1 in 1,000” becomes 10−3 10−3
Continuing in that way until we talk about 1 in 1,000,000 as 10−6 10−6
Handling Numbers (Large and Small) ▪ 71
In general, therefore, 1 divided by 10, N times, is

expressed as 10−N.
Multiplying and Dividing
Having mastered this shorthand notation, not only do very

large and very small numbers become easier to write but
the arithmetic becomes simpler. In the new notation:
100 years is written as 102
10 such periods of time is written as 101
The total number of years is therefore 100 × 10
which is, of course 1,000, or: 103
Notice how the upper-case numbers (known as the

exponents) are added when we multiply (i.e., 2 + 1 = 3).
100 bottles is written as 102
10 bottles per crate is written as 101
The total number of crates is thus
100 divided by 10 which is, of course, 10, or 101
Notice how the upper-case numbers (known as the

exponents) are subtracted when we divide (i.e., 2 − 1 = 1).
Some Examples
1,000 groups of 100 people 103 × 102 105 people

30 items for 300 years 3 101 × 3 102 9 103 item-years
500 bottles divided into 5 102 / 2 101 2.5 101 which is 25 bottles
20 crates
300 deaths from scalding in 3 102 / 6 107 0.5 10−5 which is 5 10−6
a population of 60,000,000 per head of population
Notice, in the last example, how we made the 0.5 bigger

(by ten) and the 10−5 smaller by (ten) in order to make the
first part of the number a sensible digit. In the same way:
500 104 is 5 106

0.25 106 is 2.5 105
300 10−6 is 3 10−4
0.3 10−3 is 3 10−4
Using the above rules makes handling large and small

numbers, and their arithmetic, much simpler. In spreadsheets
the 10−3 can be entered as 0.001 but, also, as 1E-3. In calcu-
lators, it is usually 1[EXP]-3, the [EXP] being one of the keys.
Familiarity with Negative Exponents
Remember:
3 10−4 is SMALLER (not BIGGER) than 3 10−3.

Handling Numbers (Large and Small) ▪ 73
Because the minus denotes the number of tens “on the

bottom line” and therefore the larger the number in the
exponent the smaller the result.
3 10−4 DIVIDED by 10 is 3 10−5 – in other words

the number in the negative exponent gets bigger
to make the resulting number smaller.
Exercise 9a Large and Small Numbers

Complete the following table:
Write the Answer as a

Number Followed by 10x
500 groups of 50 people ? people
50 men for 200 years ? man-years
500 bottles divided into 25 crates ? per crate
200 deaths from scalding in a ? chance of scalding
population of 50,000,000
3600 patients awaiting 12 beds ? patients per bed
1500 influenza fatalities in a ? chance of becoming a
population of 5,000,000 fatality
1 in 500 suffer from diabetes in a ? patients
population of 75,000
Exercise 9b Large and Small Numbers

Now try the following, without the aid of a calculator. It is
a valuable skill to be able manipulate such numbers
mentally.
Complete the following table:
Write the Answer as a Number Followed

by 10x
106 / 5 ?
4 2
6 10 / 3 10 ?
5 10−3 / 2.5 10−1 ?
9 10−1 / 3 10−2 ?
2 10−3 / 4 10−2 ?
10−4 / 5 10−5 ?
9 7
3 10 / 6 10 ?
Chapter 10
An Introduction
to Risk
The purpose of this chapter is to provide a wider per-
spective by introducing the risk of fatality from scenarios
other than those arising from medical issues. It provides a
comparison between the background occupational and
leisure related risks, to which we are exposed, and those
relating to health. The following rates and probabilities
are, by their nature, approximate estimates. References
3 and 4 (Appendix 9) also deal with this aspect.
Individual Risk of Fatality (per Exposure

to an Activity)
Table 10.1 (UK figures) shows a number of risks on the

basis of the fatality rate per single exposure to an activity.
DOI: 10.4324/9781003220138-10 75
Table 10.1 Fatality Risks per Single Exposure to an Activity

Flu Vaccination (over 65 years old) (based on 70% <3 10−7
coverage of approx 12 106over 65s with zero fatalities
at 90% confidence ∗)
Commercial aviation – one trip 1 10−7
Scuba diving (one dive) 5 10−6
Parachute/skydiving jump 6 10−6
One-hour helicopter ride (private flying) 7 10−6
General anaesthetic 1 10−5
Childbirth 8 10−5
Gastrointestinal endoscopy 1 10−4
Notes
∗Notice the inference of a probability (or less), stated at some confidence
level. This makes further use of the Chi-squared technique. The
“probability or less” is calculated as χ2 /2T where T is the aggregate
number of exposures to the risk and χ2 is obtained from n = 2(k+1)
degrees of freedom (for k occurrences) and at a probability of (1-
confidence). In the above example, χ2 is found in Appendix 7, from
n = 2(0+1) = 2 and probability 0.1, namely 4.61. Thus, 4.61/(2 × 70% ×
12 106) = <3 10−7. The same formula may be used to infer a rate, in
which case T becomes the aggregate time of exposure. The technique
is fully explained in Reference 4 ( Appendix 9).
Note, also, that an inference based on zero failures

merely establishes a maximum probability or rate (i.e.,
ceiling) below which the actual value is likely to fall. It
might well be even lower, but the inference is limited by
the size of the sample available.
An Introduction to Risk ▪ 77
Individual Risk of Fatality (per Annum)
Table 10.2 (UK figures) shows a number of risks on the

basis of the fatality rate per annum.
Table 10.2 Fatality Risks per Annum

Major accidents (e.g., Flixborough, Kegworth) 1 10−6
Carbon monoxide poisoning 1 10−6
Working in an office 1 10−6
Food poisoning 3 10−6
Homicide 1 10−5
Working in Industry (general) 1.5 10−5
Poisoning general (excluding drugs) 2 10−5
Road transport (car) 5 10−5
Asbestos related cancer (mesothelioma) 6 10−5
Suicide 1 10−4
All random accidents (non-medical) (e.g., occupational, 2 10−4
sport, transport, homicide, etc.)
Influenza/pneumonia (non-COVID) 2 10−4
Covid (UK) 2020 (50% might have died from other 5 10−4
causes)
Cancer 2 10−3
Fatal Accident Frequency (FAFR)
Comparing different activities on the basis of fatalities “per

person per annum” is not always realistic because the
“participating time” is not the same for all activities.
It is a credible approximation to say that we are all
exposed to the risks in Table 10.2 for equal amounts of
time. However, to compare motor transport and private
flying in that way would be misleading when we spend
say 500 hours per annum in the car and maybe 10 hours
per annum flying. The traditional way of comparing risks
for that type of activity is by what is known as the FAFR.
That is the number of fatalities per 100 million hours of
exposure to that activity.
Table 10.3, compiled from a number of sources, com-
pares some activities and pastimes.
Maximum Tolerable Risk
When assessing risk in industrial settings the concept of

Maximum Tolerable Risk is used. The term “Maximum
Tolerable” implies that under no circumstances would that
activity be allowed to continue if the target is not met. Any
higher risk is labelled Intolerable. It is then necessary to
demonstrate whether the target is met and, if so, by a
sufficient margin to render the cost of further risk reduc-
tion not justified. The same principle applies in medicine.
Table 10.3 FAFRs (Fatalities per 108 Hours)

Activity FAFR
Miscellaneous:
Staying at home 3
Struck by lightning 0.04
Transport:
Motorcycle 450
Rail 3
Bus 1
Pedestrian (walking) 10
Motor car 12
Canoeing 400
Commercial air (worldwide) 3
Recreation:
Horse riding 500
Rock climbing 600
Private helicopter flying 630
Private fixed wing flying 750
Gliding 670
Scuba diving 670
Swimming 1,000
The justification for expenditure on further risk reduction

is dealt with in the next section.
Typical values used in industry are shown in Table 10.4.
Table 10.4 Typical Maximum Tolerable Risk Targets

Maximum Tolerable Risk for a single fatality scenario 10−4 pa
(where exposure is voluntary (i.e., employee)
Maximum Tolerable Risk for a multi-fatality scenario 10−5 pa
(here exposure is voluntary (i.e., employee)
Maximum Tolerable Risk for a single fatality scenario 10−5 pa
(where exposure is involuntary (i.e., public)
Maximum Tolerable Risk for a multi-fatality scenario 10−6 pa
(where exposure is involuntary (i.e., public)
Variations to the above targets apply to multi-risk sce-

narios and to different numbers of fatalities. They are dealt
with in Reference 4 (See Appendix 9).
As Low as Reasonably Practicable

(ALARP)
This concept imposes the obligation to continue reducing

the risk below the Maximum Tolerable level depending
on the cost of the proposed risk reduction. In order to
apply the concept, it becomes necessary to agree a target
Cost per Life Saved. A typical value used in respect of
industrial and many other hazards is £2,000,000.
The calculation involves assessing the cost of a proposed
risk reduction measure and then calculating, based on the
predicted number of fatalities avoided, the cost for each life
saved. Where this exceeds £2,000,000 the expenditure is
deemed not be justified and where it is below £2,000,000

then that expenditure is called for.
The concept of a cost per life saved (often referred to as
cost per fatality avoided) of £2,000,000 infers that a saved
life, having avoided the hazard in question, is granted the
remainder of his/her remaining life span (say an average of
35 years). However, for the purposes of evaluating medical
expenditure, the life saved may lead to an additional span
of more restricted duration. If, for the sake of argument, a
year of extended life is obtained, then the equivalent
threshold becomes £2,000,000/35 = £57,000. It so happens
that a value of £60,000 per statistical year of life (SLY) is, in
fact, used in many health-related justifications. This SLY
needs, therefore, to be calculated as the cost of the proce-
dure divided by the anticipated number of extended years
of life.
A medical example might be the cost of a cervical smear
test which is in the order of £100. It is estimated that this
procedure saves 5,000 lives per year and leads to 5 addi-
tional years of life. Approximately 3,000,000 tests are
conducted annually. Based on those estimates the SLY is
3 106 × £100/[5 ,000 × 5] = £12 ,000
Based on the above argument, against a criterion of

£60,000, the cost of the risk reduction is well justified. The
reader may care to research the numbers used in the above
example and form his/her own opinion.
Exercise 10 SLY/Cost Per Life Saved

The following approximate figures are mooted for bowel
cancer screening. The cost of a home testing kit, and
processing, is approximately £150. 50% of kits supplied
receive a response. Ten percent of responses lead to a
diagnosis of advanced adenoma. Sixty-seven percent of
those diagnosed survive for at last 2 years. Comment on
the cost effectiveness of the screening programme.
Variable versus Constant Rates
In this chapter, we have dealt with rates and probabilities of

fatality. It is worth noting that the concept of a rate parameter,
to describe the distribution of events, implies a constant rate
(in other words random events). Were this not to be the case,
then the concept of rate becomes meaningless just as does
expressing the speed of a journey between London and
Aberdeen (say 600 miles in 10 hours) as 60 mph. In reality the
speed would have been continuously varying and the 60
mph parameter is only a metric for describing the trip. To
deal with cases where the rate is not constant, it is necessary
to make use of a more complex model, known as the Weibull
distribution. This will involve more than one parameter one
of which, known as the shape parameter, indicates whether
the rate of occurrence is increasing, constant or decreasing.
Weibull analysis is slightly beyond the scope of this book but
is dealt with fully in Reference 4 (Appendix 9) and, also, in
the COMPARE software package available from the author.
Chapter 11
A Final Word
I have tried to show, in this book, some fairly simple
techniques for drawing conclusions from quantified phy-
sical measurements. However, let us beware of the short-
sightedness which leads to a blinkered treatment of
numerical data. There is a danger in running away with
the conclusions as if they were the only factors which
impinge on the situation in question. In medical applica-
tions, this would be the patient’s overall health. I cannot
stress, too strongly, the need for a holistic approach to
health care. The role of the specialist is paramount but,
nevertheless, it is equally important to take a wide view of
the parameters involved and to seek a balance between
different treatment options.
I believe the same might be said of all scientific and
engineering disciplines.
If you have worked through the book, and attempted
the exercises, I hope you will have gained a fair grasp of
the basic principles of statistical sampling and inference. I
have attempted to explain the techniques with little more
than simple arithmetic. They can all be applied using the
tables and curves provided.
DOI: 10.4324/9781003220138-11 83
If I have removed some of the mystery and suspicion

which surrounds this subject and if you feel better
equipped to challenge numerical claims and statements
and are able to apply some of the techniques, then our
mutual efforts have not been in vain.
Appendices
Appendix 1: Manipulating Numbers in

Spreadsheets
It is useful to be familiar with the following manipulations
when using spreadsheets for statistical calculations. These
instructions relate to the spreadsheets on the following
pages. The precise keystrokes may differ slightly according
to the spreadsheet version being used. However, the
following will point the reader in the right direction.
Arithmetic Functions
Cells B1 to B6 contain 6 items of data, namely the values

2,4,6,4,6,2.
Cell B8 Adds them [=SUM(then block the 6 cells and
press return]
Cell B9 returns the average [=AVERAGE(then block the
6 cells and press return]
Cell B 10 Multiplies the 6 values together [=PRODUCT(then
block the 6 cells and press return]
Cell B11 returns the square of cell B9, namely [B9^2]
Cell B12 returns the square root of cell B11, namely
[B9^0.5]
[Note that a square root is the number raised to the
power of the reciprocal. Thus ½ =0.5]
DOI: 10.4324/9781003220138-NaN 85
86 ▪ Manipulating numbers in Spreadsheets
Cell B13 returns the standard deviation of the 6 values,

namely,
[= STDEV(then block the 6 cells and press return]
Copy–Paste
Cells D8–D12 were generated by blocking cells B8–B12,

Control C [to copy] and then click D8, Control V [to paste].
Note that the results for D8 and D10 are zero. This is be
cause EXCEL will have changed the B column designation
to D and we are now manipulating the data in D1–D6
which is, of course, zero. Cells D, D11, D12 and D13 return
#DIV/0! Because they involve a divide by cell, in the D
column, which contains 0.
Copy–Paste Special
Colum E, however, is generated by a similar process but

NOT using Control V [to paste], Instead, following the
Control C operation, block cell E8 and select paste special
and select “values”. Thus, the actual values, rather than the
amended formulae, are brought across.
Appendix 1 ▪ 87
Spreadsheet for the Chapter 2 example on

Conditional Probabilities
A useful adjunct to Chapter 2 is the use of a spreadsheet to
calculate the effectiveness of both positive and negative re
sults of tests for some condition (as for example cancer). In
cell B1, the probability of suffering from the condition in
question is entered as a percentage. This may well have been
estimated as the proportion of the population so effected.
Cells D4, D5, E4, E5 will be set according to the known re
sults pertaining to true and false positives and negatives. The
following reflects the equations given in Chapter 2
Cell D8 is D4 * B1.
Cell D9 is E4 * (1 ‒ B1).
Cell D10 is D8 + D9.
Cell F12 is D8/D10
88 ▪ Manipulating numbers in Spreadsheets
Cell D15 is E5 * (1 ‒ B1),

Cell D16 is D5 * B1,
Cell D17 is D15 + D16.
Cell F19 is D15/D17.
Regression/Correlation
A useful adjunct to Chapter 8 is the use of a spreadsheet to

generate a regression line for relating a set of two vari
ables. It also provides an estimate of the correlation
coefficient (“r”) based on a least-squares calculation.
Appendix 1 ▪ 89
The following illustration shows a spreadsheet with two

columns of data (X and Y). We proceed as follows:
Block the two columns A and B.
Insert, Chart, Scatter
Right click on any one dot and select add trend line
Right click on the line and select r2 and select display
equation
As can be seen from the above example, the correlation

is very poor. In fact, it suggests that there is no linking factor
between the two variables. Note also that the Y = mX + C
straight line equation, describing the line, is provided.
Appendix 2: Brief Glossary of Terms
À priori: Knowing in advance, rather than from
observation.
Attribute: A state that either applies or does not (e.g., a
red card or not)
Coefficient of Variation: The ratio of the standard
deviation to the mean.
Confidence: The probability that a given inference will
be found to be correct.
Correlation Coefficient: The strength of the relation
ship between two variables.
Dimensionless: Having no units of measurement.
Distribution: The scatter or spread of values presented
by a set of data.
E: The symbol that represents 10 when entering num
bers in exponential format to EXCEL and some other
packages.
Empirical: Based on actual data.
Exclusive: The situation where one attribute precludes
another (e.g., a red card and a black card).
EXP: The symbol that represents 10 when entering
numbers in exponential format into a calculator.
FAFR: Fatal Accident Frequency (usually expressed per
100 million hours of exposure/participation in an activity).
90
Appendix 2 ▪ 91
Independence: The situation where the probability of

one specified event has no influence on the probability of
another specified event.
Inference: The act of predicting the probability of
some outcome based on knowledge of the data in
question.
Mean (arithmetic): Average value based on the sum of
the values of the items divided by the number of items.
Mean (geometric): Average value based on the pro
duct of the values of the items to the power of the re
ciprocal of the number of items.
Median: The value, either side of which, 50% of the
items, in a distribution, are to be found.
Metric: A numerical measure.
Parameter: Any feature of interest which is measurable
(e.g., height, weight, blood pressure).
Population: The total number of items from which a
sample is drawn.
Probability: The proportion of items which represent
the factor in question (in other words, the ratio of the
number of occurrences to the number of trials). Note: This
might involve a sample or a complete population.
Random: A selection in which each item of a set has an
equal probability of being chosen.
Regression: The process of establishing the mathe
matical relationship between two variables.
Risk: Probability or rate of some hazardous event.
92 ▪ Brief Glossary of Terms
Skewed Distribution A non-symmetrical distribution

of values which tails (left or right).
Standard Deviation: A measure of spread given by the
square root of the sum of the squares of the differences
between each individual item and the mean.
Units: The values by which variables are described/
measured (e.g., feet, pounds)
Variable: Any measurable feature which may take a
range of values.
Variance: The square of the Standard Deviation.
Appendix 3: Normal Distribution Table
93
Appendix 4a: 0.5% & 1% Points of the F
Distribution
These F tables are “two-sided” in that they assess the sig

nificance of there being a difference between the variances, but
not distinguishing on which side they differ.
0.5% Probability
Degrees of freedom (larger std. dev.)
1 2 3 4 5 6 10 24 ∞
1 16200 20000 21600 22500 23100 23400 24200 24900 25500

Degrees of freedom (smaller std. dev.)
2 199 199 199 199 199 199 199 199 199
3 55.6 49.8 47.5 46.2 45.4 44.8 43.7 42.6 41.8
4 31.3 26.3 24.3 23.2 22.5 22.0 21.0 20.0 19.3
5 22.8 18.3 16.5 15.6 14.9 14.5 13.6 12.8 12.1
6 18.6 14.5 12.9 12.0 11.5 11.1 10.3 9.47 8.88
10 12.8 9.43 8.08 7.34 6.87 6.54 5.85 5.17 4.64
20 9.94 6.99 5.82 5.17 4.76 4.47 3.85 3.22 2.69
30 9.18 6.35 5.24 4.62 4.23 3.95 3.34 2.73 2.18
40 8.83 6.07 4.98 4.37 3.99 3.71 3.12 2.50 1.93
∞ 7.88 5.30 4.28 3.72 3.35 3.09 2.52 1.90 1.00
94
Appendix 4a ▪ 95
1% Probability
1 2 3 4 5 6 10 24 ∞
1 4050 5000 5400 5620 5760 5860 6060 6230 6370

2 98.5 99.0 99.2 99.2 99.3 99.3 99.4 99.5 99.5
3 34.1 30.8 29.5 28.7 28.2 27.9 27.2 26.6 26.1
4 21.2 18.0 16.7 16.0 15.5 15.2 14.5 13.9 13.5
5 16.3 13.3 12.1 11.4 11.0 10.7 10.1 9.47 9.02
6 13.7 10.9 9.78 9.15 8.75 8.47 7.87 7.31 6.88
10 10.0 7.56 6.55 5.99 5.64 5.39 4.85 4.33 3.91
20 8.10 5.85 4.94 4.43 4.10 3.87 3.37 2.86 2.42
30 7.56 5.39 4.51 4.02 3.70 3.47 2.98 2.47 2.01
40 7.31 5.18 4.31 3.83 3.51 3.29 2.80 2.29 1.80
∞ 6.63 4.61 3.78 3.32 3.02 2.80 2.32 1.79 1.00
Appendix 4b: 2.5% & 5% Points of the F
Distribution

nificance of there being a difference between the var
iances, but not distinguishing on which side they differ.
2.5% Probability
1 2 3 4 5 6 10 24 ∞
1 648 799 864 900 922 937 969 997 1020

2 38.5 39.0 39.2 39.2 39.3 39.3 39.4 39.5 39.5
3 17.4 16.0 15.4 15.1 14.9 14.7 14.4 14.1 13.9
4 12.2 10.6 9.98 9.60 9.36 9.20 8.84 8.51 8.26
5 10.0 8.43 7.76 7.39 7.15 6.98 6.62 6.28 6.02
6 8.81 7.26 6.60 6.23 5.99 5.82 5.46 5.12 4.85
10 6.94 5.46 4.83 4.47 4.24 4.07 3.72 3.37 3.08
20 5.87 4.46 3.86 3.51 3.29 3.13 2.77 2.41 2.09
30 5.57 4.18 3.59 3.25 3.03 2.87 2.51 2.14 1.79
40 5.42 4.05 3.46 3.13 2.90 2.74 2.39 2.01 1.64
∞ 5.02 3.69 3.12 2.79 2.57 2.41 2.05 1.64 1.00
5% Probability
1 2 3 4 5 6 10 24 ∞
1 161 199 216 225 230 234 242 249 254

2 18.5 19.0 19.2 19.2 19.3 19.3 19.4 19.5 19.5
3 10.1 9.55 9.28 9.12 9.01 8.94 8.79 8.64 8.53
4 7.71 6.94 6.59 6.39 6.26 6.16 5.96 5.77 5.63
5 6.61 5.79 5.41 5.19 5.05 4.95 4.74 4.53 4.36
6 5.99 5.14 4.76 4.53 4.39 4.28 4.06 3.84 3.67
10 4.96 4.10 3.71 3.48 3.33 3.22 2.98 2.74 2.54
20 4.35 3.49 3.10 2.87 2.71 2.60 2.35 2.08 1.84
30 4.17 3.32 2.92 2.69 2.53 2.42 2.16 1.89 1.62
40 4.08 3.23 2.84 2.61 2.45 2.34 2.08 1.79 1.51
∞ 3.84 3.00 2.60 2.37 2.21 2.10 1.83 1.52 1.00
96
Appendix 4c: 10% & 25% Points of the F
Distribution

nificance of there being a difference between the variances,
but not distinguishing on which side they differ.
10% Probability
Degrees of freedom larger (std. dev.)
1 2 3 4 5 6 10 24 ∞
1 39.9 49.5 53.6 55.8 57.2 58.2 60.2 62.0 63.3
2 8.53 9.00 9.16 9.24 9.29 9.33 9.39 9.45 9.49

3 5.54 5.46 5.39 5.34 5.31 5.28 5.23 5.18 5.13
4 4.54 4.32 4.19 4.11 4.05 4.01 3.92 3.83 3.76
5 4.06 3.78 3.62 3.52 3.45 3.40 3.30 3.19 3.10
6 3.78 3.46 3.29 3.18 3.11 3.05 2.94 2.82 2.72
10 3.29 2.92 2.73 2.61 2.52 2.46 2.32 2.18 2.06
20 2.97 2.59 2.38 2.25 2.16 2.09 1.94 1.77 1.61
30 2.88 2.49 2.28 2.14 2.05 1.98 1.82 1.64 1.46
40 2.84 2.44 2.23 2.09 2.00 1.93 1.76 1.57 1.38
∞ 2.71 2.30 2.08 1.94 1.85 1.77 1.60 1.38 1.00
25% Probability
1 2 3 4 5 6 10 24 ∞
1 5.83 7.50 8.20 8.58 8.82 8.98 9.32 9.63 9.85
2 2.57 3.00 3.15 3.23 3.28 3.31 3.38 3.43 3.48

3 2.02 2.28 2.36 2.39 2.41 2.42 2.44 2.46 2.47
4 1.81 2.00 2.05 2.06 2.07 2.08 2.08 2.08 2.08
5 1.69 1.85 1.88 1.89 1.89 1.89 1.89 1.88 1.87
6 1.62 1.76 1.78 1.79 1.79 1.78 1.77 1.75 1.74
10 1.49 1.60 1.60 1.59 1.59 1.58 1.55 1.52 1.48
20 1.40 1.49 1.48 1.47 1.45 1.44 1.40 1.35 1.29
30 1.38 1.45 1.44 1.42 1.41 1.39 1.35 1.29 1.23
40 1.36 1.44 1.42 1.40 1.39 1.37 1.33 1.26 1.19
∞ 1.32 1.39 1.37 1.35 1.33 1.31 1.25 1.18 1.00
97
Appendix 5: The t Distribution
These “t” tables are “single-sided” in that they assess the

significance of one mean being different in a specified
direction.
α is the likelihood that the value occurs by chance if the
means are indeed the same. Thus (1 – α) is the probability
that they are different.
98
Appendix 5 ▪ 99
Probability of a larger value

α 50% 40% 20% 10% 5% 2.50% 1% 0.50% 0.10%
1 1.000 1.376 3.078 6.131 12.710 25.452 63.657 >> >>
2 0.816 1.061 1.886 2.920 4.303 6.205 9.925 14.089 31.598
3 0.765 0.978 1.638 2.353 3.182 4.176 5.841 7.453 12.941
4 0.741 0.941 1.533 2.132 2.776 3.495 4.604 5.598 8.610
5 0.727 0.920 1.476 2.015 2.571 3.163 4.032 4.773 6.859
6 0.718 0.906 1.440 1.943 2.447 2.969 3.707 4.317 5.959

7 0.711 0.896 1.415 1.895 2.365 2.841 3.499 4.029 5.405
8 0.706 0.889 1.397 1.860 2.306 2.752 3.355 3.832 5.041
9 0.703 0.883 1.383 1.833 2.262 2.685 3.250 3.690 4.871
10 0.770 0.879 1.372 1.812 2.228 2.634 3.169 3.581 4.587
11 0.697 0.876 1.363 1.796 2.201 2.593 3.106 3.497 4.437

12 0.695 0.873 1.356 1.782 2.197 2.560 3.055 3.428 4.318
13 0.694 0.870 1.350 1.771 2.160 2.533 3.012 3.372 4.221
14 0.692 0.878 1.354 1.761 2.145 2.510 2.977 3.326 4.140
15 0.691 0.866 1.341 1.753 2.131 2.490 2.847 3.286 4.073
16 0.690 0.865 1.337 1.746 2.120 2.473 2.921 3.252 4.015

17 0.689 0.863 1.333 1.740 2.110 2.458 2.898 3.222 3.965
18 0.688 0.862 1.330 1.734 2.101 2.445 2.878 3.197 3.922
19 0.688 0.861 1.328 1.729 2.093 2.433 2.861 3.174 3.883
Degrees of freedom
20 0.687 0.860 1.325 1.275 2.086 2.423 2.845 3.153 3.550
21 0.686 0.859 1.323 1.721 2.080 2.414 2.831 3.135 3.819

22 0.686 0.858 1.321 1.717 2.074 2.406 2.819 3.119 3.792
23 0.685 0.858 1.319 1.714 2.069 2.398 2.807 3.104 3.762
24 0.685 0.857 1.318 1.711 2.064 2.391 2.797 3.090 3.745
25 0.684 0.856 1.316 1.708 2.060 2.385 2.787 3.078 3.725
26 0.684 0.856 1.315 1.706 2.056 2.379 2.779 3.067 3.707

27 0.684 0.855 1.314 1.703 2.052 2.373 2.771 3.056 3.690
28 0.683 0.855 1.313 1.701 2.048 2.368 2.763 3.047 3.674
29 0.683 0.854 1.311 1.699 2.045 2.364 2.756 3.038 3.659
30 0.683 0.854 1.310 1.697 2.042 2.360 2.750 3.030 3.646
35 0.682 0.852 1.306 1.690 2.030 2.342 2.724 2.996 3.591

40 0.681 0.851 1.303 1.684 2.021 2.329 2.704 2.971 3.551
45 0.680 0.850 1.301 1.680 2.014 2.319 2.690 2.952 3.520
50 0.680 0.849 1.299 1.676 2.008 2.310 2.678 2.937 3.496
55 0.679 0.849 1.297 1.673 2.004 2.304 2.669 2.925 3.476
60 0.679 0.848 1.296 1.671 2.000 2.299 2.660 2.915 3.460

70 0.678 0.847 1.294 1.667 1.994 2.290 2.648 2.899 3.435
80 0.678 0.847 1.293 1.665 1.989 2.284 2.638 2.887 3.416
90 0.678 0.846 1.291 1.662 1.986 2.279 2.631 2.878 3.402
100 0.677 0.846 1.290 1.661 1.982 2.276 2.625 2.871 3.390
120 0.672 0.845 1.289 1.658 1.980 2.270 2.617 2.860 3.373
∞ 0.675 0.842 1.282 1.645 1.960 2.241 2.576 2.807 3.291
Note » signifies – too large to consider.

Appendix 6: Cumulative Poisson Curves
100
Appendix 7: Percentage Points of the Chi
Squared Distribution
These Chi-squared tables are “single-sided” in that they

assess the significance of there being a difference in a
specified direction.
Probability that χ2 > or = to the stated value
0.999 0.99 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.01
1 <0.001 <0.001 0.02 0.06 0.15 0.28 0.46 0.70 1.10 1.64 2.71 6.63
2 0.00 0.02 0.21 0.45 0.71 1.02 1.39 1.83 2.41 3.22 4.61 9.21
3 0.02 0.12 0.58 1.00 1.42 1.87 2.37 2.95 3.67 4.64 6.25 11.30
4 0.09 0.30 1.06 1.65 2.19 2.75 3.36 4.04 4.88 5.99 7.78 13.30
5 0.21 0.55 1.61 2.34 3.00 3.66 4.35 5.13 6.06 7.29 9.24 15.10
6 0.38 0.87 2.20 3.07 3.83 4.57 5.35 6.21 7.23 8.56 10.60 16.80
7 0.60 1.24 2.83 3.82 4.67 5.49 6.35 7.28 8.38 9.80 12.00 18.50
8 0.86 1.65 3.49 4.59 5.53 6.42 7.34 8.35 9.52 11.00 13.40 20.10
Degrees of freedom
9 1.15 2.09 4.17 5.38 6.39 7.36 8.34 9.41 10.70 12.20 14.70 21.70
10 1.48 2.56 4.87 6.18 7.27 8.30 9.34 10.50 11.80 13.40 16.00 23.20
15 3.48 5.23 8.55 10.30 11.70 13.00 14.30 15.70 17.30 19.30 22.30 30.60
20 5.92 8.26 12.40 14.60 16.30 17.80 19.30 21.00 22.80 25.00 28.40 37.60
25 8.65 11.50 16.50 18.90 20.90 22.60 24.30 26.10 28.20 30.70 34.40 44.30
30 11.60 15.00 20.60 23.40 25.50 27.40 29.30 31.30 33.50 36.30 40.30 50.90
35 14.70 18.50 24.80 27.80 30.20 32.30 34.30 36.50 38.90 41.80 46.10 57.30
40 17.90 22.20 29.10 32.30 34.90 37.10 39.30 41.60 44.20 47.30 51.80 63.70
45 21.30 25.90 33.40 36.90 39.60 42.00 44.30 46.80 49.50 52.70 57.50 70.00
50 24.70 29.70 37.70 41.40 44.30 46.90 49.30 51.90 54.70 58.20 63.20 76.20
75 42.80 49.50 59.80 64.50 68.10 71.30 74.30 77.50 80.90 85.10 91.10 106.40
100 61.90 70.10 82.40 87.90 92.10 95.80 99.30 103.00 107.00 112.00 118.50 135.80
101
Appendix 8: Answers to Exercises
Exercise 1 Manipulating Probabilities

Having dark hair is 80% likely; Having blue eyes is 25%
likely; Being bald is 5% likely; Being taller than 1.8 m is
50% likely
1. Having dark hair and blue eyes?

80% × 25% = 20%
2. Having dark hair or blue eyes?
Not 80% + 25% = 105% (more than 1!!!!)
Remember it is Pa+ Pb − PaPb
So, 80% + 25% − (80% × 25%) = 85%
3. Having dark hair and being bald?
Not 80% × 5%!!!!
One cannot be both (they are mutually exclusive)
4. Having dark hair and blue eyes and being taller than
1.8 m?
80% × 25% × 50% = 10%
5. Having dark hair or blue eyes or being taller than 1.8
m?
1 − (1–80%) × (1–25%) × (1–50%) = 92.5%
6. Having dark hair and blue eyes or being bald?
Being bald is mutually exclusive, thus we add the 50%
without subtracting PaPb.
Thus (80% × 25%) + 5% = 25%
7. Having neither blue eyes nor not being bald?
Not having blue eyes is (1–25%) = 75%
Not being bald is (1–5%) = 95%
Thus, having neither is 75% × 95% = 71.25%
102
Appendix 8 ▪ 103
Exercise 2 Obtaining the Mean and Standard

Deviation
The totals for the columns no in group, total years,

squared difference have used the =SUM(….) Microsoft
EXCEL Function. The mean can be obtained either by
dividing the 119 total by the cell containing 14 OR by
using the =AVERAGE(….) Microsoft EXCEL Function.
The variance is the 23.5 cell divided by the 14 cell and
the standard deviation by taking the square root
(i.e., cell^0.5).
104 ▪ Answers to Exercises
Exercise 3 Mean and Coefficient of Variation

Values of mmol/L
The mean has been calculated as 6.73.

Below each daily reading is the square of the difference
between that value and the mean. By taking the square
root of the average, shown in the right-hand column, the
standard deviation is obtained.
This is then expressed as a percentage of the mean.
Thus, it could be inferred that 68% of the values all fall
between + or – 2.8% of the mean value. In other words,
the patient’s blood sugar levels are significantly consistent.
Exercise 4 Median Life Expectancy
Thus, the median is 5 ½ years and the chance of

surviving beyond 6 years is 28%.
Appendix 8 ▪ 105
Exercise 5 Comparing Two Distributions

The following spreadsheet has been used to calculate the
two means and standard deviations and then to calculate
the values F and t.
There being 6 items in each sample of data, there are 5
degrees of freedom.
From the F tables, the 2.75 is (between 10% and 25%)

thus approximately 15% likely to be by chance. In other
words, we can only be about 85% sure that the difference
is significant and that the Group 2 results were more
consistent.
From the t tables the 1.14 is about 30% likely to be by
chance. In other words, we can only be about 70% sure
that the difference is significant and that the treatment is
effective.
Exercise 6 Attributes Using Poisson Curves

A practice covers 2,000 potential patients, and it is known
that the incidence of stroke in the UK is approximately
115,000 pa (population 67 million). Thus, the frequency is
= 115k/67 m = 1.7 10−3 pa.
Therefore, m = 1.7 10−3 × 2,000 = 3.4
From the Appendix 6 curves, the probability that the
practice will need to treat:
0 cases = 3%
1 or less case = 15% thus 1 case 15%−3%=12%
2 or less cases = 33% thus 2 cases 33−15%=18%
3 or less cases = 55% thus 3 cases 55%−33%=22%
Appendix 8 ▪ 107
Exercise 7 Chi-Squared
It might therefore be argued that (for this group of

patients only) the insulin alone made little or no significant
improvement to the condition and that it is just 60% likely
that the combined treatment resulted in an improvement.
The outcome would therefore be encouraging but not
conclusive.
Exercise 8 Correlation
If the spreadsheet has been properly constructed, then the
r = 0.32 should have been obtained. The formula for the
cell D33 is shown below and represents the formula in
Chapter 8
=F28/(G28*H28)^0.5
Cells B28–H28 are the sums of the rows 2–25.
Cells B30–C30 are the averages of the rows 2–25.
The data entered thus indicates a very poor correlation

between the systolic and the diastolic. In fact, r2 becomes
about 0.1.
Appendix 8 ▪ 109
Exercise 9a Large and Small Numbers
Write Each as a Write the Answer

Number as a Number
Followed by 10x Followed by 10x
500 groups of 50 5 102 × 5 101 25 103 = 2.5 104
people people
50 men for 200 years 5 101 × 2 102 10 103 = 104
manyears
500 bottles divided 5 102 / 2.5 101 2 101 i.e., 20 per
into 25 crates crate
200 deaths from 2 102 / 5 107 0.4 10−5 = 4 10−6
scalding in a popu chance of
lation of 50,000,000 scalding
3600 patients awaiting 3.6 103 / 12 0.3 103 i.e., 300
12 beds patients per bed
1500 influenza fatal 1.5 103 / 5 106 0.3 10−3 3 10−4 =
ities in a population chance of be
of 5,000,000 coming a fatality
1 in 500 suffer from (1/500) × 75 103 150 100 = 150
diabetes in a popu = 2 10−3 × patients*
lation of 75,000 75 103
*Notice how +3 plus −3 = 0 and thus 100 which is one.

Exercise 9b Large and Small Numbers
Write the Hints About the Thought Process

Answer
as a
Number
Followed
by 10x
106 / 5 2 105 5 on the bottom means 2 on the
top but reduce the 6 (by 1) to 5.
6 104 / 3 102 2 102 6/3 = 2 and exponent 4 − 2 = ex
ponent 2
5 10−3 / 2.5 10−1 2 10−2 5/2.5 = 2 and exponent − 3 take
away −1 (i.e., add 1) gives − 2
9 10−1 / 3 10−2 3 101 = 30 9/3 = 3 and exponent −1 take
away −2 (i.e., add 2) gives +1
2 10−3 / 4 10−2 5 10−2 2/4 means 5 on the top BUT
reduce the exponent by 1. Thus,
− 3 − 1 − (− 2) = − 2
10−4 / 5 10−5 2 100 = 2 5 on the bottom means 2 on the
top but reduce the
exponent by 1.
Thus, − 4 − 1 − (− 5) = 0
3 109 / 6 107 5 101 = 50 3/6=5 but reduce the exponent by
1. Thus, 9 − 7 = 2. Then 2 − 1 = 1
Appendix 8 ▪ 111
Exercise 10 SLY/Cost Per Life Saved

The cost per patient screened, given the 50% take-up, is
2 × £150 = £300.
It could be argued that 67% of 10% of recipients enjoy
an additional 2 years of life.
Thus, the SLY based on the 2 years additional life
expectancy:
£300/(0.67 × 0.1 × 2) = £2 ,2 40
Given the above assumptions, against a criterion of

£60,000 per SLY it might be argued that the programme is
highly cost effective.
The reader may care to research these assumptions and
construct his/her own argument.
Appendix 9: Further Reading
1. Facts from Figures by M J Moroney (ISBN:

9780140202366).
2. Statistical Methods, 8th edition William G. Cochran
George W. Snedecor ISBN 978-8126551385.
3. Technis Guidelines T810 Fatal accident frequency
statistics FAFRs – A summary, from david.smith@
technis.org.uk
4. Reliability, Maintainability and Risk, 9th Edition (2017)
Smith DJ, (Elsevier) ISBN 978008102010-4.
112
Index
A coefficient of variation (COV),

23–25, 90
à priori statement of probability, 7, 90
conditional probabilities, 13–16
as low as reasonably practicable
confidence, 2, 21, 22–23, 90
(ALARP), 80–82
likelihood of, 22
attributes, 3, 90
refers to pro ability that inference
binomial, 49–51
is true, 22
compared to variables, 3, 17
consistency, 3, 4
poisson distribution, 51–53, 106
control charts, 47–48
testing for significance, 55–59
correlation, 4, 61–65, 108
average, 1
false, 65–66
correlation coefficient, 61, 90
B spreadsheets, 88–89
Bayes Theorem, 16
binomial attribute, 49–51
D
binomial expansion, 51
data as integer, 21
data comparisons, 3
C
data variability, 2–3
Chi squared distribution, 55, 58–59, dimensionless, 8, 90
75, 101, 107 distribution, 90
113
114 ▪ Index
Chi-squared, 55, 58–59, 75, I

101, 107
comparison of two, 31–32, 110 independence, 91
event, 82 inferences, 21, 22, 91
Fisher’s F, 29, 94–97 double-sided, 32
skewed, 25–26, 92 goodness of fit to normal
t, 101–102 distribution curve, 22
unbiased (symmetrical), 21, 26 single-sided, 32
Weibull, 82
L
E
linear, 2
E, 90 logarithmic scales, 43–45
empirical statement of probability, 7, 90 compared to linear scales, 43–44
exclusive, 8, 11, 90
EXP, 90
M
F
mean (arithmetic), 1, 91
fatal accident frequency (FAFR), comparison of, 31–32
77–78, 91 formula, 19
Fisher’s F distribution, 29, 94–97 as measure of central tendency
(or average), 18
spread of values and, 2
G value, 3
mean (geometric), 27–28, 91
geometric mean, 27–28
mean and coefficient of variation,
graphs, 38–40
25, 104
technique, 2
mean deviation, 20, 103
visual plot (linear), 1
median life expectancy, 26–27, 104
grouped data, 20
medians, 25–27, 91
metric, 91
H moving averages, 45–47
multiplicity rule, 9
histograms (bar charts), 33–38 mutually exclusive events, 11–12
Index ▪ 115
N
R
negative exponents, 72–74
normal (or Gaussian) distribution, random sampling, 5, 91
18, 22, 26, 29, 32, 38, relevant factors in, 5
51–52, 68 ratios, 10
comparison of two distributions, regression, 61–65, 91
29, 30–32 equation, 63–64
establishment of, 66–68 spreadsheets for line
table, 94 generation, 88–89
numbers: big, 69–70, 73–74, regression coefficient, 68
114–115 risk of fatality, 75, 91
division, 71–72 fatal accident frequency
multiplication, 71–72 (FAFR), 77–78
as positive and negative powers individual per annum, 77
of ten, 5–6, 69 individual per exposure to an
small, 70, 73–74, 110 activity, 75–76
as low as reasonably practical
(ALARP), 80–82
P maximum tolerable, 78–80
variable vs. constant rates, 82
parameter, 91
poisson curves, 53
attributes using, 53, 111 S
cumulative, 104
poisson distribution, 51–53 scatter diagrams, 40, 51
population, 91 semi-quantified metrics, 17–18
power of ten, 69 as variables, 17
probability, 7–8, 91 significance testing, 55–59
à priori statement of, 7 spread (distribution) of values, 2–3
combination of, 8–11 spread (distribution) of
concept of, 8 variable, 3–4
conditional, 13–16, 87–88 spreadsheets (for statistical
either or both, 10 calculations), 85–89
empirical statement of, 7 arithmetic functions, 85–86
false and true negatives, conditional probabilities, 87–88
14–15, 16 copy-paste, 86
false and true positives, correlation coefficient, 88–89
13–14, 15–16 regression line generation, 88–89
manipulation, 12, 102 standard deviation, 19, 20, 92, 103
116 ▪ Index
comparison of, 29–31 semi-quantified metrics as, 17–18

formula, 19–20 spread of, 3–4
technique to describe strength
between two, 61–65
T variance, 92
visual plot (linear graph), 1
t distribution, 98–99 visual presentation of data, 33–41
control charts, 47–48
extrapolation as pitfall, 41–43
U graphs, 38–40
units, 92 histograms, 33–38
logarithmic scales as pitfall, 43–45
moving averages as pitfall, 45–47
V suppression as pitfall, 41
variables, 3, 92
compared to attributes, 3, 17 W
continuous, 17
expressed in terms of continuous Weibull distribution, 82
measurement, 17
relationship to other variables, 4

David J. Smith - Basic Statistical Techniques For Medical and Other Professionals - A Course in Statistics To Assist in Interpreting Numerical Data-Productivity Press (2021)

Uploaded by

Copyright:

Available Formats

David J. Smith - Basic Statistical Techniques For Medical and Other Professionals - A Course in Statistics To Assist in Interpreting Numerical Data-Productivity Press (2021)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

David J. Smith - Basic Statistical Techniques For Medical and Other Professionals - A Course in Statistics To Assist in Interpreting Numerical Data-Productivity Press (2021)

Uploaded by

Copyright:

Available Formats

Basic Statistical

Techniques for Medical

Dr. David J. Smith

ISBN: 978-1-032-11495-8 (hbk)

1 Why One Needs Statistical Techniques .............1

3 Dealing with Variables....................................17

6 Dealing with Attributes...................................49

I have had the pleasure of knowing David Smith for nearly

knowledge of basic statistical techniques. It is also aimed at

The purpose of this book is to provide members of the

My sincere thanks to Sam Samuel, without whose patience

Dr David J Smith BSc, PhD, FIET, FCGI, HonFSaRS,

By the same author:

We are surrounded by numerical data in every walk of life.

I may be accused of oversimplifying the subject but I am

Now let us examine the meaning of the word Probability.

the population and NOT by experimentally drawing cards

Why One Needs

7.5, 7.2, 8, 6, 5.6, 5, 7, 5.5, 5, 5.5

The average can easily be calculated as 6.2 although it may

An average (we use the word MEAN in statis­

Figure 1.1 Simple graph of a variable against time

Both can be extremely helpful.

levels of 5 and 8 mmol/L in the earlier example. Chapter 4 will

Variables and Attributes

At this point, it will be useful to distinguish between

The blood sugar readings illustrated in Figure 1.1 are dis­

measurement in question. In the blood sugar example, a

Another problem, which statistics will help to unravel, is

Much of statistical analysis consists of drawing conclusions

Random sampling is the ideal way of collecting data

Very Large and Very Small Numbers

In many fields, data involves numbers in the hundreds of

in a hundred thousand and so on. “One in a hundred

Probability and Its

An estimate of the probability of an event is given by the

In both cases, the concept of probability is derived from a

Assume that the probability of a person, selected at

Figure 2.1 Overlapping events

The shaded (overlapping) portion of the diagram re­

If Pa (the probability of event A occurring) is 0.8

Then the probability that both event A and event B will

In other words, 0.8 proportion of the 0.7 proportion of the

The above equations assume independence between the

might not be independent as, for example, a genetic factor,

Examining this in terms of ratios, 0.8 of events will turn out to

1−Pa is the probability of event A NOT occurring

It follows that for n events the Probability of one or more

Mutually Exclusive Events

Let us now examine the case of mutually exclusive events.

Figure 2.2 Mutually exclusive events

and for multiple events, the probability of observing either

In the playing card example Pa = 0.25 Pb = 0.5 and

Exercise 1 Manipulating Probabilities

Having dark hair is 80% likely.

What is the probability of the following:

1. Having dark hair and blue eyes?

The following example illustrates a state of affairs in which

Patient Has Patient Has No

An average (we use the word MEAN in statis

The blood sugar readings illustrated in Figure 1.1 are dis

The shaded (overlapping) portion of the diagram re

In the case of testing positive the chances of a result, de

Inference means that statements based on the mathe

The geometric mean of two numbers, X and Y, is calcu

A factor known as F (from Fisher’s F distribution) is cal