Chapter 1

Chapter 1

Introduction, descriptive statistics, R

and data visualization
Chapter 1

Contents

1 Introduction, descriptive statistics, R and data visualization

1.1 What is Statistics - a primer . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Statistics at DTU Compute . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Statistics - why, what, how? . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.1 Measures of centrality . . . . . . . . . . . . . . . . . . . . . 8
1.4.2 Measures of variability . . . . . . . . . . . . . . . . . . . . . 12
1.4.3 Measures of relation: correlation and covariance . . . . . . 16
1.5 Introduction to R and RStudio . . . . . . . . . . . . . . . . . . . . . 20
1.5.1 Console and scripts . . . . . . . . . . . . . . . . . . . . . . . 21
1.5.2 Assignments and vectors . . . . . . . . . . . . . . . . . . . 21
1.5.3 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . 22
1.5.4 Use of R in the course and at the exam . . . . . . . . . . . . 24
1.6 Plotting, graphics - data visualisation . . . . . . . . . . . . . . . . 26
1.6.1 Frequency distributions and the histogram . . . . . . . . . 26
1.6.2 Cumulative distributions . . . . . . . . . . . . . . . . . . . 28
1.6.3 The box plot and the modified box plot . . . . . . . . . . . 29
1.6.4 The Scatter plot . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.6.5 Bar plots and Pie charts . . . . . . . . . . . . . . . . . . . . 36
1.6.6 More plots in R? . . . . . . . . . . . . . . . . . . . . . . . . 38
1.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Glossaries 42

Acronyms 43
Chapter 1 1.1 WHAT IS STATISTICS - A PRIMER 1

This is the first chapter in the eight-chapter DTU Introduction to Statistics book.
It consists of eight chapters:

1. Introduction, descriptive statistics, R and data visualization

2. Probability and simulation

3. Statistical analysis of one and two sample data

4. Statistics by simulation

5. Simple linear regression

6. Multiple linear regression

7. Analysis of categorical data

8. Analysis of variance (analysis of multi-group data)

In this first chapter the idea of statistics is introduced together with some of the
basic summary statistics and data visualization methods. The software used
throughout the book for working with statistics, probability and data analysis is
the open source environment R. An introduction to R is included in this chapter.

1.1 What is Statistics - a primer

To catch your attention we will start out trying to give an impression of the
importance of statistics in modern science and engineering.

In the well respected New England Journal of medicine a millennium editorial on

the development of medical research in a thousand years was written:

EDITORIAL: Looking Back on the Millennium in Medicine, N Engl J Med, 342:42-

49, January 6, 2000, NEJM200001063420108.

They came up with a list of 11 points summarizing the most important devel-
opments for the health of mankind in a millennium:

• Elucidation of human anatomy and physiology

• Discovery of cells and their substructures
• Elucidation of the chemistry of life
• Application of statistics to medicine
• Development of anaesthesia
Chapter 1 1.1 WHAT IS STATISTICS - A PRIMER 2

• Discovery of the relation of microbes to disease

• Elucidation of inheritance and genetics
• Knowledge of the immune system
• Development of body imaging
• Discovery of antimicrobial agents
• Development of molecular pharmacotherapy

The reason for showing the list here is pretty obvious: one of the points is Ap-
plication of Statistics to Medicine! Considering the other points on the list, and
what the state of medical knowledge was around 1000 years ago, it is obviously
a very impressive list of developments. The reasons for statistics to be on this
list are several and we mention two very important historical landmarks here.
Quoting the paper:

"One of the earliest clinical trials took place in 1747, when James Lind treated 12
scorbutic ship passengers with cider, an elixir of vitriol, vinegar, sea water, oranges
and lemons, or an electuary recommended by the ship’s surgeon. The success of the
citrus-containing treatment eventually led the British Admiralty to mandate the provi-
sion of lime juice to all sailors, thereby eliminating scurvy from the navy." (See also
James_Lind).

Still today, clinical trials, including the statistical analysis of the outcomes, are
taking place in massive numbers. The medical industry needs to do this in
order to find out if their new developed drugs are working and to provide doc-
umentation to have them accepted for the World markets. The medical industry
is probably the sector recruiting the highest number of statisticians among all
sectors. Another quote from the paper:

"The origin of modern epidemiology is often traced to 1854, when John Snow demon-
strated the transmission of cholera from contaminated water by analyzing disease rates
among citizens served by the Broad Street Pump in London’s Golden Square. He ar-
rested the further spread of the disease by removing the pump handle from the polluted
well." (See also John_Snow_(physician)).

Still today, epidemiology, both human and veterinarian, maintains to be an ex-

tremely important field of research (and still using a lot of statistics). An im-
portant topic, for instance, is the spread of diseases in populations, e.g. virus
spreads like Ebola and others.

Actually, today more numbers/data than ever are being collected and the amounts
are still increasing exponentially. One example is Internet data, that internet
companies like Google, Facebook, IBM and others are using extensively. A
quote from New York Times, 5. August 2009, from the article titled “For To-
Chapter 1 1.2 STATISTICS AT DTU COMPUTE 3

day’s Graduate, Just One Word: Statistics” is:

“I keep saying that the sexy job in the next 10 years will be statisticians," said Hal
Varian, chief economist at Google. ‘and I’m not kidding.’ ”

The article ends with the following quote:

“The key is to let computers do what they are good at, which is trawling these massive
data sets for something that is mathematically odd,” said Daniel Gruhl, an I.B.M. re-
searcher whose recent work includes mining medical data to improve treatment. “And
that makes it easier for humans to do what they are good at - explain those anomalies.”

1.2 Statistics at DTU Compute

At DTU Compute at the Technical University of Denmark statistics is used,

taught and researched mainly within four research sections:

• Statistics and Data Analysis

• Dynamical Systems
• Image Analysis & Computer Graphics
• Cognitive Systems

Each of these sections have their own focus area within statistics, modelling
and data analysis. On the master level it is an important option within DTU
Compute studies to specialize in statistics of some kind on the joint master pro-
gramme in Mathematical Modelling and Computation (MMC). And a Statisti-
cian is a well-known profession in industry, research and public sector institu-
tions.

The high relevance of the topic of statistics and data analysis today is also illus-
trated by the extensive list of ongoing research projects involving many and di-
verse industrial partners within these four sections. Neither society nor indus-
try can cope with all the available data without using highly specialized peo-
ple in statistical techniques, nor can they cope and be internationally competi-
tive without continuously further developing these methodologies in research
projects. Statistics is and will continue to be a relevant, viable and dynamic
field. And the amount of experts in the field continues to be small compared
to the demand for experts, hence obtaining skills in statistics is for sure a wise
career choice for an engineer. Still for any engineer not specialising in statistics,
a basic level of statistics understanding and data handling ability is crucial for
the ability to navigate in modern society and business, which will be heavily
influenced by data of many kinds in the future.
Chapter 1 1.3 STATISTICS - WHY, WHAT, HOW? 4

1.3 Statistics - why, what, how?

Often in society and media, the word statistics is used simply as the name for
a summary of some numbers, also called data, by means of a summary table
and/or plot. We also embrace this basic notion of statistics, but will call such
basic data summaries descriptive statistics or explorative statistics. The meaning
of statistics goes beyond this and will rather mean “how to learn from data in an
insightful way and how to use data for clever decision making”, in short we call this
inferential statistics . This could be on the national/societal level, and could be
related to any kind of topic, such as e.g. health, economy or environment, where
data is collected and used for learning and decision making. For example:

• Cancer registries
• Health registries in general
• Nutritional databases
• Climate data
• Macro economic data (Unemployment rates, GNP etc. )
• etc.

The latter is the type of data that historically gave name to the word statistics. It
originates from the Latin ‘statisticum collegium’ (state advisor) and the Italian
word ‘statista’ (statesman/politician). The word was brought to Denmark by
the Gottfried Achenwall from Germany in 1749 and originally described the
processing of data for the state, see also History_of_statistics.

Or it could be for industrial and business applications:

• Is machine A more effective than machine B?
• How many products are we selling on different markets?
• Predicting wind and solar power for optimizing energy systems
• Do we produce at the specified quality level?
• Experiments and surveys for innovative product development
• Drug development at all levels at e.g. Novo Nordisk A/S or other phar-
maceutical companies
• Learning from "Big Data"
• etc.

In general, it can be said say that we learn from data by analysing the data
with statistical methods. Therefore statistics will in practice involve mathematical
Chapter 1 1.3 STATISTICS - WHY, WHAT, HOW? 5

modelling, i.e. using some linear or non-linear function to model the particular
phenomenon. Similarly, the use of probability theory as the concept to describe
randomness is extremely important and at the heart of being able to “be clever”
in our use of the data. Randomness express that the data just as well could have
come up differently due to the inherent random nature of the data collection
and the phenomenon we are investigating.

Probability theory is in its own right an important topic in engineering relevant

applied mathematics. Probability based modelling is used for e.g. queuing sys-
tems (queuing for e.g. servers, websites, call centers etc.), for reliability mod-
elling, and for risk analysis in general. Risk analysis encompasses a vast di-
versity of engineering fields: food safety risk (toxicological and/or allergenic),
environmental risk, civil engineering risks, e.g. risk analysis of large building
constructions, transport risk, etc. The present material focuses on the statistical
issues, and treats probability theory at a minimum level, focusing solely on the
purpose of being able to do proper statistical inference and leaving more elabo-
rate probability theory and modelling to other texts.

There is a conceptual frame for doing statistical inference: in Statistical inference

the observed data is a sample, that is (has been) taken from a population. Based
on the sample, we try to generalize to (infer about) the population. Formal
definitions of what the sample and the population is are given by:

Definition 1.1 Sample and population

• An observational unit is the single entity about which information is

sought (e.g. a person)
• An observational variable is a property which can be measured on the
observational unit (e.g. the height of a person)
• The statistical population consists of the value of the observational vari-
able for all observational units (e.g. the heights of all people in Den-
mark)
• The sample is a subset of the statistical population, which has been cho-
sen to represent the population (e.g. the heights of 20 persons in Den-
mark).

See also the illustration in Figure 1.1.

This is all a bit abstract at this point. And likely adding to the potential confu-
sion about this is the fact that the words population and sample will have a “less
Chapter 1 1.3 STATISTICS - WHY, WHAT, HOW? 6

(Infinite) Statistical population

Sample
{ x1 , x2 , . . . , x n }
Randomly
selected

Statistical
Mean Sample mean
Inference
µ x̄

Figure 1.1: Illustration of statistical population and sample, and statistical in-
ference. Note that the bar on each person indicates that the it is the height (the
observational variable) and not the person (the observational unit), which are
the elements in the statistical population and the sample. Notice, that in all
analysis methods presented in this text the statistical population is assumed to
be very large (or infinite) compared to the sample size.

precise” meaning when used in everyday language. When they are used in a
statistical context the meaning is very specific, as given by the definition above.
Let us consider a simple example:

Example 1.2

The following study is carried out (actual data collection): the height of 20 persons
in Denmark is measured. This will give us 20 values x1 , . . . , x20 in cm. The sample
is then simply these 20 values. The statistical population is the height values of all
people in Denmark. The observational unit is a person.

The meaning of sample in statistics is clearly different from how a chemist or

medical doctor would use the word, where a sample would be the actual sub-
stance in e.g. the petri dish. Within this book, when using the word sample, then
it is always in the statistical meaning i.e. a set of values taken from a statistical
population.

With regards to the meaning of population within statistics the difference to the
Chapter 1 1.4 SUMMARY STATISTICS 7

everyday meaning is less obvious: but note that the statistical population in the
example is defined to be the height values of people, not actually the people.
Had we measured the weights instead the statistical population would be quite
different. Also later we will realize that statistical populations in engineering
contexts can refer to many other things than populations as in a group of or-
ganisms, hence stretching the use of the word beyond the everyday meaning.
From this point: population will be used instead of statistical population in order
to simplify the text.

The population in a given situation will be linked with the actual study and/or
experiment carried out - the data collection procedure sometimes also denoted
the data generating process. For the sample to represent relevant information
about the population it should be representative for that population. In the ex-
ample, had we only measured male heights, the population we can say any-
thing about would be the male height population only, not the entire height
population.

A way to achieve a representative sample is that each observation (i.e. each

value) selected from the population, is randomly and independently selected of
each other, and then the sample is called a random sample.

1.4 Summary statistics

The descriptive part of studying data maintains to be an important part of statis-

tics. This implies that it is recommended to study the given data, the sample,
by means of descriptive statistics as a first step, even though the purpose of a full
statistical analysis is to eventually perform some of the new inferential tools
taught in this book, that will go beyond the pure descriptive part. The aims of
the initial descriptive part are several, and when moving to more complex data
settings later in the book, it will be even more clear how the initial descriptive
part serves as a way to prepare for and guide yourself in the subsequent more
formal inferential statistical analysis.

The initial part is also called an explorative analysis of the data. We use a number
of summary statistics to summarize and describe a sample consisting of one or
two variables:

• Measures of centrality:
– Mean
– Median
– Quantiles
Chapter 1 1.4 SUMMARY STATISTICS 8

• Measures of “spread”:
– Variance
– Standard deviation
– Coefficient of variation
– Inter Quartile Range (IQR)
• Measures of relation (between two variables):
– Covariance
– Correlation

One important point to notice is that these statistics can only be calculated for
the sample and not for the population - we simply don’t know all the values
in the population! But we want to learn about the population from the sample.
For example when we have a random sample from a population we say that the
sample mean (x̄) is an estimate of the mean of the population, often then denoted
µ, as illustrated in Figure 1.1.

Remark 1.3
Notice, that we put ’sample’ in front of the name of the statistic, when it is
calculated for the sample, but we don’t put ’population’ in front when we
refer to it for the population (e.g. we can think of the mean as the true mean).

HOWEVER we don’t put sample in front of the name every time it should
be there! This is to keep the text simpler and since traditionally this is not
strictly done, for example the median is rarely called the sample median,
even though it makes perfect sense to distinguish between the sample me-
dian and the median (i.e. the population median). Further, it should be
clear from the context if the statistic refers to the sample or the population,
when it is not clear then we distinguish in the text. Most of the way we do
distinguish strictly for the mean, standard deviation, variance, covariance and
correlation.

1.4.1 Measures of centrality

The sample mean is a key number that indicates the centre of gravity or cen-
tring of the sample. Given a sample of n observations x1 , . . . , xn , it is defined as
Chapter 1 1.4 SUMMARY STATISTICS 9

follows:

Definition 1.4 Sample mean

The sample mean is the sum of observations divided by the number of ob-
servations
1 n
x̄ =
n i∑
xi . (1-1)
=1

Sometimes this is refereed to as the average.

The median is also a key number indicating the center of sample (note that to
be strict we should call it ’sample median’, see Remark 1.3 above). In some
cases, for example in the case of extreme values or skewed distributions, the
median can be preferable to the mean. The median is the observation in the
middle of the sample (in sorted order). One may express the ordered observa-
tions as x(1) , . . . , x(n) , where then x(1) is the smallest of all x1 , . . . , xn (also called
the minimum) and x(n) is the largest of all x1 , . . . , xn (also called the maximum).

Definition 1.5 Median

Order the n observations x1 , . . . , xn from the smallest to largest:
x(1) , . . . , x(n) . The median is defined as:
n +1
• If n is odd the median is the observation in position 2 :

Q 2 = x ( n +1 ) . (1-2)
2

• If n is even the median is the average of the two observations in posi-

tions n2 and n+ 2
2 :

x ( n ) + x ( n +2 )
Q2 =
2 2
. (1-3)
2
The reason why it is denoted with Q2 is explained below in Definition 1.8.
Chapter 1 1.4 SUMMARY STATISTICS 10

Example 1.6 Student heights

A random sample of the heights (in cm) of 10 students in a statistics class was

168 161 167 179 184 166 198 187 191 179 .

The sample mean height is

1
x̄ = (168 + 161 + 167 + 179 + 184 + 166 + 198 + 187 + 191 + 179) = 178.
10
To find the sample median we first order the observations from smallest to largest

x (1) x (2) x (3) x (4) x (5) x (6) x (7) x (8) x (9) x(10)
.
161 166 167 168 179 179 184 187 191 198

Note that having duplicate observations (like e.g. two of 179) is not a problem - they
all just have to appear in the ordered list. Since n = 10 is an even number the median
becomes the average of the 5th and 6th observations
x( n2 ) + x( n+2 ) x (5) + x (6) 179 + 179
2
= = = 179.
2 2 2
As an illustration, let’s look at the results if the sample did not include the 198 cm
height, hence for n = 9

1
x̄ = (168 + 161 + 167 + 179 + 184 + 166 + 187 + 191 + 179) = 175.78.
9
then the median would have been

x( n+1 ) = x(5) = 179.

2

This illustrates the robustness of the median compared to the sample mean: the
sample mean changes a lot more by the inclusion/exclusion of a single “extreme”
measurement. Similarly, it is clear that the median does not depend at all on the
actual values of the most extreme ones.

The median is the point that divides the observations into two halves. It is of
course possible to find other points that divide into other proportions, they are
called quantiles or percentiles (note, that this is actually the sample quantile or
sample percentile, see Remark 1.3).
Chapter 1 1.4 SUMMARY STATISTICS 11

Definition 1.7 Quantiles and percentiles

The p quantile also called the 100p% quantile or 100p’th percentile, can be
defined by the following procedure: a

1. Order the n observations from smallest to largest: x(1) , . . . , x(n)

2. Compute pn

3. If pn is an integer: average the pn’th and ( pn + 1)’th ordered observa-

tions. Then the p quantile is
 
q p = x(np) + x(np+1) /2 (1-4)

4. If pn is a non-integer: take the “next one” in the ordered list. Then the
p’th quantile is

q p = x(dnpe) , (1-5)

where dnpe is the ceiling of np, that is, the smallest integer larger than
np

a There exist several other formal definitions. To obtain this definition of quan-
tiles/percentiles in R use quantile(. . . , type=2). Using the default in R is also a perfectly
valid approach - just a different one.

Often calculated percentiles are the so-called quartiles (splitting the sample in
quarters, i.e. 0%, 25%, 50%, 75% and 100%):

• q0 , q0.25 , q0.50 , q0.75 and q1

Note that the 0’th percentile is the minimum (smallest) observation and the
100’th percentile is the maximum (largest) observation. We have specific names
for the three other quartiles:

Definition 1.8 Quartiles

Q1 = q0.25 = “lower quartile” = “0.25 quantile” = “25’th percentile”
Q2 = q0.50 = “median” = “0.50 quantile” = “50’th percentile”
Q3 = q0.75 = “upper quartile” = “0.75 quartile” = “75’th percentile”
Chapter 1 1.4 SUMMARY STATISTICS 12

Example 1.9 Student heights

Using the n = 10 sample from Example 1.6 and the ordered data table from there,
let us find the lower and upper quartiles (i.e. Q1 and Q3 ), as we already found
Q2 = 179.

First, the Q1 : with p = 0.25, we get that np = 2.5 and we find that

Q1 = x(d2.5e) = x(3) = 167,

and since n · 0.75 = 7.5, the upper quartile becomes

Q3 = x(d7.5e) = x(8) = 187.

We could also find the 0’th percentile

q0 = min( x1 , . . . , xn ) = x(1) = 161,

and the 100’th percentile

q1 = max( x1 , . . . , xn ) = x(10) = 198.

Finally, 10’th percentile (i.e. 0.10 quantile) is

x (1) + x (2) 161 + 166
q0.10 = = = 163.5,
2 2
since np = 1 for p = 0.10.

1.4.2 Measures of variability

A crucial aspect to understand when dealing with statistics is the concept of

variability - the obvious fact that not everyone in a population, nor in a sample,
will be exactly the same. If that was the case they would all equal the mean
of the population or sample. But different phenomena will have different de-
grees of variation: An adult (non dwarf) height population will maybe spread
from around 150 cm up to around 210 cm with very few exceptions. A kitchen
scale measurement error population might span from −5 g to +5 g. We need a
way to quantify the degree of variability in a population and in a sample. The
most commonly used measure of sample variability is the sample variance or
its square root, called the sample standard deviation:
Chapter 1 1.4 SUMMARY STATISTICS 13

Definition 1.10 Sample variance

The sample variance of a sample x1 , . . . , xn is the sum of squared differences
from the sample mean divided by n − 1
n
1
s2 = ( xi − x̄ )2 .
n − 1 i∑
(1-6)
=1

Definition 1.11 Sample standard deviation

The sample standard deviation is the square root of the sample variance
s
√ n
1
s = s2 = ( xi − x̄ )2 .
n − 1 i∑
(1-7)
=1

The sample standard deviation and the sample variance are key numbers of
absolute variation. If it is of interest to compare variation between different
samples, it might be a good idea to use a relative measure - most obvious is the
coefficient of variation:

Definition 1.12 Coefficient of variation

The coefficient of variation is the sample standard deviation seen relative to
the sample mean
s
V= . (1-8)
x̄

We interpret the standard deviation as the average absolute deviation from the mean
or simply: the average level of differences, and this is by far the most used measure
of spread. Two (relevant) questions are often asked at this point (it is perfectly
fine if you didn’t wonder about them by now and you might skip the answers
and return to them later):
Chapter 1 1.4 SUMMARY STATISTICS 14

Remark 1.13

Question: Why not actually compute directly what the interpretation is

stating, which would be: n1 ∑in=1 | xi − x̄ |?

Answer: This is indeed an alternative, called the mean absolute deviation, that
one could use. The reason for most often measuring “mean deviation”
NOT by the Mean Absolute Deviation statistic, but rather by the sample
standard deviation s, is the so-called theoretical statistical properties of
the sample variance s2 . This is a bit early in the material for going into
details about this, but in short: inferential statistics is heavily based
on probability considerations, and it turns out that it is theoretically
much easier to put probabilities related to the sample variance s2 on
explicit mathematical formulas than probabilities related to most other
alternative measures of variability. Further, in many cases this choice
is in fact also the optimal choice in many ways.

Remark 1.14

Question: Why divide by n − 1 and not n in the formulas of s and s2 ?

(which also appears to fit better with the stated interpretation)

Answer: The sample variance s2 will most often be used as an estimate of

the (true but unknown) population variance σ2 , which is the average
of ( xi − µ)2 in the population. In doing that, one should ideally com-
pare each observation xi with the population mean, usually called µ.
However, we do not know µ and instead we use x̄ in the computation
of s2 . In doing so, the squared differences ( xi − x̄ )2 that we compute in
this way will tend to be slightly smaller than those we ideally should
have used: ( xi − µ)2 (as the observations themselves were used to find
x̄ so they will be closer to x̄ than to µ). It turns out, that the correct way
to correct for this is by dividing by n − 1 instead of n.

Spread in the sample can also be described and quantified by quartiles:

Chapter 1 1.4 SUMMARY STATISTICS 15

Definition 1.15 Range

The range of the sample is

Range = Maximum − Minimum = Q4 − Q0 = x(n) − x(1) . (1-9)

The Inter Quartile Range (IQR) is the middle 50% range of data defined as

IQR = q0.75 − q0.25 = Q3 − Q1 . (1-10)

Example 1.16 Student heights

Consider again the n = 10 data from Example 1.6. To find the variance let us com-
pute the n = 10 differences to the mean, that is ( xi − 178)

-10 -17 -11 1 6 -12 20 9 13 1.

So, if we square these and add them up we get

10
∑ (xi − x̄)2 = 102 + 172 + 112 + 12 + 62 + 122 + 202 + 92 + 132 + 12 = 1342.
i =1

Therefore the sample variance is

1
s2 = 1342 = 149.1,
9
and the sample standard deviation is

s = 12.21.

We can interpret this as: people are on average around 12 cm away from the mean
height of 178 cm. The Range and Inter Quartile Range (IQR) are easily found from
the ordered data table in Example 1.6 and the earlier found quartiles in Example 1.9

Range = maximum − minimum = 198 − 161 = 37,

IQR = Q3 − Q1 = 187 − 167 = 20.

Hence 50% of all people (in the sample) lie within 20 cm.

Note, that the standard deviation in the example has the physical unit cm,
Chapter 1 1.4 SUMMARY STATISTICS 16

whereas the variance has cm2 . This illustrates the fact that the standard de-
viation has a more direct interpretation than the variance in general.

1.4.3 Measures of relation: correlation and covariance

When two observational variables are available for each observational unit, it
may be of interest to quantify the relation between the two, that is to quantify
how the two variables co-vary with each other, their sample covariance and/or
sample correlation.

Example 1.17 Student heights and weights

In addition to the previously given student heights we also have their weights (in
kg) available

Heights ( xi ) 168 161 167 179 184 166 198 187 191 179
.
Weights (yi ) 65.5 58.3 68.1 85.7 80.5 63.4 102.6 91.4 86.7 78.9

The relation between weights and heights can be illustrated by the so-called scatter-
plot, cf. Section 1.6.4, where e.g. weights are plotted versus heights:

7
100

8
90

4 9
Weight

y = 78.1 5
80

10
70

3
1
6
x = 178
60

2
160 170 180 190
Height

Each point in the plot corresponds to one student - here illustrated by using the
observation number as plot symbol. The (expected) relation is pretty clear now -
different wordings could be used for what we see:

• Weights and heights are related to each other

Chapter 1 1.4 SUMMARY STATISTICS 17

• Higher students tend to weigh more than smaller students

• There is an increasing pattern from left to right in the "point cloud”
• If the point cloud is seen as an (approximate) ellipse, then the ellipse clearly is
horizontally upwards ”tilted”.
• Weights and heights are (positively) correlated to each other

The sample covariance and sample correlation coefficients are a summary statis-
tics that can be calculated for two (related) sets of observations. They quantify
the (linear) strength of the relation between the two. They are calculated by
combining the two sets of observations (and the means and standard deviations
from the two) in the following ways:

Definition 1.18 Sample covariance

The sample covariance is
n
1
s xy = ( xi − x̄ ) (yi − ȳ) .
n − 1 i∑
(1-11)
=1

Definition 1.19 Sample correlation

The sample correlation coefficient is
n   
s xy
1 xi − x̄ yi − ȳ
r= ∑ = , (1-12)
n − 1 i =1 sx sy s x · sy

where s x and sy is the sample standard deviation for x and y respectively.

When xi − x̄ and yi − ȳ have the same sign, then the point ( xi , yi ) give a positive
contribution to the sample correlation coefficient and when they have opposite
signs the point give a negative contribution to the sample correlation coefficient,
as illustrated here:
Chapter 1 1.4 SUMMARY STATISTICS 18

Example 1.20 Student heights and weights

The sample means are found to be

x̄ = 178 and ȳ = 78.1.

Using these we can show how each student deviate from the average height and
weight (these deviations are exactly used for the sample correlation and covariance
computations)

Student 1 2 3 4 5 6 7 8 9 10
Height ( xi ) 168 161 167 179 184 166 198 187 191 179
Weight (yi ) 65.5 58.3 68.1 85.7 80.5 63.4 102.6 91.4 86.7 78.9
( xi − x̄ ) -10 -17 -11 1 6 -12 20 9 13 1
(yi − ȳ) -12.6 -19.8 -10 7.6 2.4 -14.7 24.5 13.3 8.6 0.8
( xi − x̄ )(yi − ȳ) 126.1 336.8 110.1 7.6 14.3 176.5 489.8 119.6 111.7 0.8

Student 1 is below average on both height and weight (−10 and − 12.6). Student
10 is above average on both height and weight (+1 and + 0.8).s

The sample covariance is then given by the sum of the 10 numbers in the last row of
the table
1
s xy = (126.1 + 336.8 + 110.1 + 7.6 + 14.3 + 176.5 + 489.8 + 119.6 + 111.7 + 0.8)
9
1
= · 1493.3
9
= 165.9

And the sample correlation is then found from this number and the standard devia-
tions

s x = 12.21 and sy = 14.07.

(the details of the sy computation is not shown). Thus we get the sample correlation
as
165.9
r= = 0.97.
12.21 · 14.07

Note how all 10 contributions to the sample covariance are positive in the ex-
ample case - in line with the fact that all observations are found in the first
and third quadrants of the scatter plot (where the quadrants are defined by the
sample means of x and y). Observations in second and fourth quadrant would
contribute with negative numbers to the sum, hence such observations would
be from students with below average on one feature while above average on the
Chapter 1 1.4 SUMMARY STATISTICS 19

other. Then it is clear that: had all students been like that, then the covariance
and the correlation would have been negative, in line with a negative (down-
wards) trend in the relation.

We can state (without proofs) a number of properties of the sample correlation

r:

Remark 1.21 Properties of the sample correlation, r

• r is always between −1 and 1: −1 ≤ r ≤ 1

• r measures the degree of linear relation between x and y
• r = ±1 if and only if all points in the scatterplot are exactly on a line
• r > 0 if and only if the general trend in the scatterplot is positive
• r < 0 if and only if the general trend in the scatterplot is negative

The sample correlation coefficient measures the degree of linear relation be-
tween x and y, which imply that we might fail to detect non-linear relationships,
illustrated in the following plot of four different point clouds and their sample
correlations:
Chapter 1 1.5 INTRODUCTION TO R AND RSTUDIO 20

r ≈ 0.95 r ≈ −0.5
1.2

1
0.8

0
y

y
0.4

-1
-2
0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x

r≈0 r≈0
2
1

0.8
0
y

y
0.4
-1
-2

0.0
-3

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x

The sample correlation in both the bottom plots are close to zero, but as we see
from the plot this number itself doesn’t imply that there no relation between y
and x - which clearly is the case in the bottom right and highly non-linear case.

Sample covariances and correlation are closely related to the topic of linear re-
gression, treated in Chapter 5 and 6 , where we will treat in more detail how
we can find the line that could be added to such scatter-plots to describe the re-
lation between x and y in a different (but related) way, as well as the statistical
analysis used for this.

1.5 Introduction to R and RStudio

The program R is an open source software for statistics that you can download
to your own laptop for free. Go to http://mirrors.dotsrc.org/cran/ and se-
lect your platform (Windows, Mac or Linux) and follow instructions to install.

RStudio is a free and open source integrated development environment (IDE)

for R. You can run it on your desktop (Windows, Mac or Linux) or even over
the web using RStudio Server. It works as (an extended) alternative to running R
in the basic way through a terminal. This will be used in the course. Download
it from http://www.rstudio.com/ and follow installation instructions. To use
Chapter 1 1.5 INTRODUCTION TO R AND RSTUDIO 21

the software, you only need to open RStudio (R will then be used by RStudio for
carrying out the calculations).

1.5.1 Console and scripts

Once you have opened RStudio, you will see a number of different windows.
One of them is the console. Here you can write commands and execute them by
hitting Enter. For instance:

> # Add two numbers in the console

> 2+3

[1] 5

In the console you cannot go back and change previous commands

and neither can you save your work for later. To do this you need to
write a script. Go to File->New->R Script. In the script you can write
a line and execute it in the console by hitting Ctrl+Enter (Windows)
or Cmd+Enter (Mac). You can also mark several lines and execute them
all at the same time.

1.5.2 Assignments and vectors

If you want to assign a value to a variable, you can use = or <-. The latter is the
preferred by R-users, so for instance:

> # Assign the value 3 to y

> y <- 3

It is often useful to assign a set of values to a variable like a vector. This is done
with the function c (short for concatenate):

# Concatenate numbers to a vector

x <- c(1, 4, 6, 2)
x

[1] 1 4 6 2
Chapter 1 1.5 INTRODUCTION TO R AND RSTUDIO 22

Use the colon :, if you need a sequence, e.g. 1 to 10:

> # A sequence from 1 to 10

> x <- 1:10
> x

[1] 1 2 3 4 5 6 7 8 9 10

You can also make a sequence with a specific step-size different from 1

> # Sequence with specified steps

> x <- seq(0, 1, by=0.1)
> x

[1] 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

If you are in doubt of how to use a certain function, the help page can be opened
by typing ? followed by the function, e.g. ?seq.

If you know Matlab then this document Hiebeler-matlabR.pdf can be

very helpful.

1.5.3 Descriptive statistics

All the summary statistics measures presented in Section 1.4 can be found as
functions or part of functions in R:

• mean(x) - mean value of the vector x

• var(x) - variance
• sd(x) - standard deviation
• median(x) - median
• quantile(x,p) - finds the pth quantile. p can consist of several different
values, e.g. quantile(x,c(0.25,0.75)) or quantile(x,c(0.25,0.75), type=2)
• cov(x, y) - the covariance of the vectors x and y
• cor(x, y) - the correlation
Chapter 1 1.5 INTRODUCTION TO R AND RSTUDIO 23

Please again note that the words quantiles and percentiles are used interchange-
ably - they are essentially synonyms meaning exactly the same, even though the
formal distinction has been clarified earlier.

Example 1.22 Summary statistics in R

Consider again the n = 10 data from Example 1.6. We can read these data into R
and compute the sample mean and sample median as follows:

# Sample Mean and Median

x <- c(168, 161, 167, 179, 184, 166, 198, 187, 191, 179)
mean(x)

[1] 178

median(x)

[1] 179

The sample variance and sample standard deviation are found as follows:

# Sample variance and standard deviation

var(x)

[1] 149.1

sqrt(var(x))

[1] 12.21

sd(x)

[1] 12.21

The sample quartiles can be found by using the quantile function as follows:

# Sample quartiles
quantile(x, type=2)

0% 25% 50% 75% 100%

161 167 179 187 198
Chapter 1 1.5 INTRODUCTION TO R AND RSTUDIO 24

The option “type=2” makes sure that the quantiles found by the function is found
using the definition given in Definition 1.7. By default, the quantile function would
use another definition (not detailed here). Generally, we consider this default choice
just as valid as the one explicitly given here, it is merely a different one. Also the
quantile function has an option called “probs” where any list of probability values
from 0 to 1 can be given. For instance:

# Sample quantiles 0%, 10%,..,90%, 100%:

quantile(x, probs=seq(0, 1, by=0.10), type=2)

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
161.0 163.5 166.5 168.0 173.5 179.0 184.0 187.0 189.0 194.5 198.0

1.5.4 Use of R in the course and at the exam

You should bring your laptop with R installed with you to the teaching activity
and to the exam. We will need access to the so-called probability distributions
to do statistical computations, and the values of these distributions are not oth-
erwise part of the written material: These probability distributions are part of
many different software, also Excel, but it is part of the syllabus to be able to
work with these within R.

Apart from access to these probability distributions, the R-software is used in

three ways in our course

1. As a pedagogical learning tool: The random variable simulation tools in-

built in R enables the use of R as a way to illustrate and learn the principles
of statistical reasoning that are the main purposes of this course.

2. As a pocket calculator substitute - that is making R calculate ”manually”

- by simple routines - plus, minus, squareroot etc. whatever needs to be
calculated, that you have identified by applying the right formulas from
the proper definitions and methods in the written material.

3. As a ”probability calculus and statistical analysis machine” where e.g.

with some data fed into it, it will, by inbuilt functions and procedures
do all relevant computations for you and present the final results in some
overview tables and plots.

We will see and present all three types of applications of R during the course.
For the first type, the aim is not to learn how to use the given R-code itself
Chapter 1 1.5 INTRODUCTION TO R AND RSTUDIO 25

but rather to learn from the insights that the code together with the results of
applying it is providing. It will be stated clearly whenever an R-example is of
this type. Types 2 and 3 are specific tools that should be learned as a part of the
course and represent tools that are explicitly relevant in your future engineering
activity. It is clear that at some point one would love to just do the last kind
of applications. However, it must be stressed that even though the program is
able to calculate things for the user, understanding the details of the calculations
must NOT be forgotten - understanding the methods and knowing the formulas
is an important part of the syllabus, and will be checked at the exam.

Remark 1.23 BRING and USE pen and paper PRIOR to R

For many of the exercises that you are asked to do it will not be possible
to just directly identify what R-command(s) should be used to find the re-
sults. The exercises are often to be seen as what could be termed “problem
mathematics” exercises. So, it is recommended to also bring and use pen
and paper to work with the exercises to be able to subsequently know how
to finally finish them by some R-calculations. (If you adjusted yourself to
some digital version of ”pen-and-paper”, then this is fine of course.)

Remark 1.24 R is not a substitute for your brain activity in this

course!
The software R should be seen as the most fantastic and easy computa-
tional companion that we can have for doing statistical computations that
we could have done ”manually”, if we wanted to spend the time doing
it. All definitions, formulas, methods, theorems etc. in the written mate-
rial should be known by the student, as should also certain R-routines and
functions.

A good question to ask yourself each time that you apply en inbuilt R-function
is: ”Would I know how to make this computation ”manually”?”. There are few
exceptions to this requirement in the course, but only a few. And for these the
question would be: ”Do I really understand what R is computing for me now?”
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 26

1.6 Plotting, graphics - data visualisation

A really important part of working with data analysis is the visualisation of the
raw data, as well as the results of the statistical analysis – the combination of
the two leads to reliable results. Let us focus on the first part now, which can
be seen as being part of the explorative descriptive analysis also mentioned in
Section 1.4. Depending on the data at hand different types of plots and graphics
could be relevant. One can distinguish between quantitative vs. categorical data.
We will touch on the following type of basic plots:

• Quantitative data:
– Frequency plots and histograms
– box plots
– cumulative distribution
– Scatter plot (xy plot)
• Categorical data:
– Bar charts
– Pie charts

1.6.1 Frequency distributions and the histogram

The frequency distribution is the count of occurrences of values in the sample

for different classes using some classification, for example in intervals or by
some other property. It is nicely depicted by the histogram, which is a bar plot
of the occurrences in each classes.
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 27

Example 1.25 Histogram in R

Consider again the n = 10 sample from Example 1.6.

# A histogram of the heights

hist(x)
4
3
Frequency
2
1
0

160 170 180 190 200

x

The default histogram uses equidistant interval widths (the same width for all
intervals) and depicts the raw frequencies/counts in each interval. One may
change the scale into showing what we will learn to be densities by dividing the
raw counts by n and the interval width, i.e.

"Interval count"
.
n · ("Interval width")

By plotting the densities a density histogram also called the empirical density
the area of all the bars add up to 1:
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 28

Example 1.26 Empirical density in R

# A density histogram or empirical density of the heights

hist(x, prob=TRUE, col="red", nclass=8)
0.06
0.04
Density
0.02
0.00

160 170 180 190 200

x

The R-function hist makes some choice of the number of classess based on
the number of observations - it may be changed by the user option nclass as
illustrated here, although the original choice seems better in this case due to the
very small sample.

1.6.2 Cumulative distributions

The cumulative distribution can be visualized simply as the cumulated relative

frequencies either across classes, as also used in the histogram, or individual
data points, which is then called the empirical cumulative distribution function:
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 29

Example 1.27 Cumulative distribution plot in R

# Empirical cumulative distribution plot

plot(ecdf(x), verticals=TRUE)
1.0
0.8
0.6
Fb( x )
0.4
0.2
0.0

160 170 180 190 200

x

The empirical cumulative distribution function Fn is a step function with jumps

i/n at observation values, where i is the number of identical(tied) observations
at that value.

For observations ( x1 , x2 , . . . , xn ), Fn ( x ) is the fraction of observations less or

equal to x, that mathematically can be expressed as

1
Fn ( x ) = ∑ . (1-13)
j where x ≤ x
n
j

1.6.3 The box plot and the modified box plot

The so-called box plot in its basic form depicts the five quartiles (min, Q1 , me-
dian, Q3 , max) with a box from Q1 to Q3 emphasizing the Inter Quartile Range
(IQR):
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 30

Example 1.28 Box plot in R

# A basic box plot of the heights (range=0 makes it "basic")

boxplot(x, range=0, col="red", main="Basic box plot")
# Add the blue text
text(1.3, quantile(x), c("Minimum","Q1","Median","Q3","Maximum"),
col="blue")

Basic box plot

Maximum
190

Q3
180

Median
170

Q1
Minimum
160

In the modified box plot the whiskers only extend to the min. and max. obser-
vation if they are not too far away from the box: defined to be 1.5 × IQR. Obser-
vations further away are considered as extreme observations and will be plotted
individually - hence the whiskers extend from the smallest to the largest obser-
vation within a distance of 1.5 × IQR of the box (defined as either 1.5 × IQR
larger than Q3 or 1.5 × IQR smaller than Q1 ).
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 31

Example 1.29 Box plot in R

If we add an extreme observation, 235 cm, to the heights sample and make the mod-
ified box plot - the default in R- and the basic box plot, then we have:

# Add an extreme value and box plot

boxplot(c(x, 235), col="red", main="Modified box plot")
boxplot(c(x, 235), col="red", main="Basic box plot", range=0)

Modified box plot Basic box plot

Maximum Maximum
220

220
200

200
Q3 Q3
180

180

Median Median

Q1 Q1
Minimum Minimum
160

160

Note that since there was no extreme observations among the original 10 observa-
tions, the two ”different” plots would be the same if we didn’t add the extreme 235
cm observation.

The box plot hence is an alternative to the histogram in visualising the distribu-
tion of the sample. It is a convenient way of comparing distributions in different
groups, if such data is at hand.

Example 1.30 Box plot in R

This example shows some ways of working with R to illustrate data.

In another statistics course the following heights of 17 female and 23 male students
were found:

Males 152 171 173 173 178 179 180 180 182 182 182 185
185 185 185 185 186 187 190 190 192 192 197
Females 159 166 168 168 171 171 172 172 173 174 175 175
175 175 175 177 178
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 32

The two modified box plots of the distributions for each gender can be generated by
a single call to the boxplot function:

# Box plot with two groups

Males <- c(152, 171, 173, 173, 178, 179, 180, 180, 182, 182, 182, 185,
185 ,185, 185, 185 ,186 ,187 ,190 ,190, 192, 192, 197)
Females <-c(159, 166, 168 ,168 ,171 ,171 ,172, 172, 173, 174 ,175 ,175,
175, 175, 175, 177, 178)
boxplot(list(Males, Females), col=2:3, names=c("Males", "Females"))
190
180
170
160

Males Females

At this point, it should be noted that in real work with data using R, one would
generally not import data into R by explicit listings in an R-script as here. This
only works for very small data sets. Usually the data is imported from some-
where else, e.g. from a spread sheet exported in a .csv (comma separated values)
format as shown here:
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 33

Example 1.31 Read and explore data in R

The gender grouped student heights data used in Example 1.30 is avail-
able as a .csv-file via http://www2.compute.dtu.dk/courses/introstat/data/
studentheights.csv. The structure of the data file, as it would appear in a spread
sheet program (e.g. LibreOffice Calc or Excel) is two columns and 40+1 rows includ-
ing a header row:

1 Height Gender
2 152 male
3 171 male
4 173 male
. . .
. . .
24 197 male
25 159 female
26 166 female
27 168 female
. . .
. . .
39 175 female
40 177 female
41 178 female

The data can now be imported into R with the read.table function:

# Read the data (note that per default sep="," but here semicolon)
studentheights <- read.table("studentheights.csv", sep=";", dec=".",
header=TRUE, stringsAsFactors=TRUE)

The resulting object studentheights is now a so-called data.frame, which is the

class used for such tables in R. There are some ways of getting a quick look at what
kind of data is really in a data set:
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 34

# Have a look at the first 6 rows of the data

head(studentheights)

Height Gender
1 152 male
2 171 male
3 173 male
4 173 male
5 178 male
6 179 male

# Get an overview
str(studentheights)

’data.frame’: 40 obs. of 2 variables:

$ Height: int 152 171 173 173 178 179 180 180 182 182 ...
$ Gender: Factor w/ 2 levels "female","male": 2 2 2 2 2 2 2 2 2 2 ...

# Get a summary of each column/variable in the data

summary(studentheights, quantile.type=2)

Height Gender
Min. :152.0 female:17
1st Qu.:172.5 male :23
Median :177.5
Mean :177.9
3rd Qu.:185.0
Max. :197.0

For quantitative variables we get the quartiles and the mean from summary. For cat-
egorical variables we see the category frequencies. A data structure like this is com-
monly encountered (and often the only needed) for statistical analysis. The gender
grouped box plot can now be generated by:
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 35

# Box plot for each gender

boxplot(Height ~ Gender, data=studentheights, col=2:3)

190
180
Height
170
160

female male
Gender

The R-syntax Height ~ Gender with the tilde symbol “~” is one that we will use a
lot in various contexts such as plotting and model fitting. In this context it can be
understood as “Height is plotted as a function of Gender”.

1.6.4 The Scatter plot

The scatter plot can be used for two quantitative variables. It is simply one
variable plotted versus the other using some plotting symbol.

Example 1.32 Explore data included in R

Now we will use a data set available as part of R itself. Both base R and many add-
on R-packages include data sets, which can be used for testing and practising. Here
we will use the mtcars data set. If you write:

# See information about the mtcars data

?mtcars

you will be able to read the following as part of the help info:

“The data was extracted from the 1974 Motor Trend US magazine, and comprises fuel con-
sumption and 10 aspects of automobile design and performance for 32 automobiles (1973-74
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 36

models). A data frame with 32 observations on 11 variables. Source: Henderson and Velle-
man (1981), Building multiple regression models interactively. Biometrics, 37, 391-411.”

Let us plot the gasoline use, (mpg=miles pr. gallon), versus the weight (wt):

# To make 2 plots
par(mfrow=c(1,2))
# First the default version
plot(mtcars$wt, mtcars$mpg, xlab="wt", ylab="mpg")
# Then a nicer version
plot(mpg ~ wt, xlab="Car Weight (1000lbs)", data=mtcars,
ylab="Miles pr. Gallon", col=factor(am),
main="Inverse fuel usage vs. size")
# Add a legend to the plot
legend("topright", c("Automatic transmission","Manual transmission"),
col=c("black","red"), pch=1, cex=0.7)

Inverse fuel usage vs. size

Automatic transmission
Manual transmission
30

30
Miles pr. Gallon
25

25
mpg
20

20
15

15
10

10

2 3 4 5 2 3 4 5
wt Car Weight (1000lbs)

In the second plot call we have used the so-called formula syntax of R, that was
introduced above for the grouped box plot. Again, it can be read: “mpg is plotted
as a function of wt”. Note also how a color option, col=factor(am), can be used to
group the cars with and without automatic transmission, stored in the data column
am in the data set.

1.6.5 Bar plots and Pie charts

All the plots described so far were for quantitative variables. For categorical
variables the natural basic plot would be a bar plot or pie chart visualizing the
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 37

relative frequencies in each category.

Example 1.33 Bar plots and Pie charts in R

For the gender grouped student heights data used in Example 1.30 we can plot the
gender distribution by:

# Barplot
barplot(table(studentheights$Gender), col=2:3)
20
15
10
5
0

female male

# Pie chart
pie(table(studentheights$Gender), cex=1, radius=1)

female

male
Chapter 1 1.6 PLOTTING, GRAPHICS - DATA VISUALISATION 38

1.6.6 More plots in R?

A good place for getting more inspired on how to do easy and nice plots in R is:
http://www.statmethods.net/.
Chapter 1 1.7 EXERCISES 39

1.7 Exercises

Exercise 1.1 Infant birth weight

In a study of different occupational groups the infant birth weight was recorded
for randomly selected babies born by hairdressers, who had their first child.
The following table shows the weight in grams (observations specified in sorted
order) for 10 female births and 10 male births:

Females (x) 2474 2547 2830 3219 3429 3448 3677 3872 4001 4116
Males (y) 2844 2863 2963 3239 3379 3449 3582 3926 4151 4356

Solve at least the following questions a)-c) first “manually” and then by the
inbuilt functions in R. It is OK to use R as alternative to your pocket calculator
for the “manual” part, but avoid the inbuilt functions that will produce the
results without forcing you to think about how to compute it during the manual
part.

a) What is the sample mean, variance and standard deviation of the female
births? Express in your own words the story told by these numbers. The
idea is to force you to interpret what can be learned from these numbers.

b) Compute the same summary statistics of the male births. Compare and
explain differences with the results for the female births.

c) Find the five quartiles for each sample — and draw the two box plots with
pen and paper (i.e. not using R.)

d) Are there any “extreme” observations in the two samples (use the modified
box plot definition of extremness)?

e) What are the coefficient of variations in the two groups?

Chapter 1 1.7 EXERCISES 40

Exercise 1.2 Course grades

To compare the difficulty of 2 different courses at a university the following

grades distributions (given as number of pupils who achieved the grades) were
registered:

Course 1 Course 2 Total

Grade 12 20 14 34
Grade 10 14 14 28
Grade 7 16 27 43
Grade 4 20 22 42
Grade 2 12 27 39
Grade 0 16 17 33
Grade -3 10 22 32
Total 108 143 251

a) What is the median of the 251 achieved grades?

b) What are the quartiles and the IQR (Inter Quartile Range)?

Exercise 1.3 Cholesterol

In a clinical trial of a cholesterol-lowering agent, 15 patients’ cholesterol (in

mmol L−1 ) was measured before treatment and 3 weeks after starting treatment.
Data is listed in the following table:

Patient 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Before 9.1 8.0 7.7 10.0 9.6 7.9 9.0 7.1 8.3 9.6 8.2 9.2 7.3 8.5 9.5
After 8.2 6.4 6.6 8.5 8.0 5.8 7.8 7.2 6.7 9.8 7.1 7.7 6.0 6.6 8.4

a) What is the median of the cholesterol measurements for the patients before
treatment, and similarly after treatment?

b) Find the standard deviations of the cholesterol measurements of the pa-

tients before and after treatment.
Chapter 1 1.7 EXERCISES 41

c) Find the sample covariance between cholesterol measurements of the pa-

tients before and after treatment.

d) Find the sample correlation between cholesterol measurements of the pa-

tients before and after treatment.

e) Compute the 15 differences (Dif = Before − After) and do various sum-

mary statistics and plotting of these: sample mean, sample variance, sam-
ple standard deviation, boxplot etc.

f) Observing such data the big question is whether an average decrease in

cholesterol level can be “shown statistically”. How to formally answer
this question is presented in Chapter 3, but consider now which summary
statistics and/or plots would you look at to have some idea of what the
answer will be?

Exercise 1.4 Project start

a) Go to CampusNet and take a look at the first project and read the project
page on the website for more information (02323.compute.dtu.dk/projects
or 02402.compute.dtu.dk/projects). Follow the steps to import the data
into R and get started with the explorative data analysis.
Chapter 1 Glossaries 42

Glossaries

Box plot [Box plot] The so-called boxplot in its basic form depicts the five quar-
tiles (min, Q1 , median, Q3 , max) with a box from Q1 to Q3 emphasizing
the IQR 26, 29–32, 34, 36

Categorical data [Kategorisk data] A variable is called categorical if each ob-

servation belongs to one of a set of categories 1, 26

Class The frequency distribution of the data for a certain grouping of the data
26, 28

Correlation [Korrelation] The sample correlation coefficient are a summary statis-

tic that can be calculated for two (related) sets of observations. It quantifies
the (linear) strength of the relation between the two. See also: Covariance
16–20, 22

Covariance [Kovarians] The sample covariance coefficient are a summary statis-

tic that can be calculated for two (related) sets of observations. It quantifies
the (linear) strength of the relation between the two. See also: Correlation
16–20, 22

Descriptive statistics [Beskrivende statistik] Descriptive statistics, or explorative

statistics, is an important part of statistics, where the data is summarized
and described 1, 4, 7

Empirical cumulative distribution [Empirisk fordeling] The empirical cumu-

lative distribution function Fn is a step function with jumps i/n at obser-
vation values, where i is the number of identical observations at that value
28, 29

Frequency [Frekvens] How frequent data is observed. The frequency distribu-

tion of the data for a certain grouping is nicely depicted by the histogram,
which is a barplot of either raw frequencies or for some number of classes
26–28, 34, 37
Chapter 1 Glossaries 43

Histogram [Histogram] The default histogram uses the same width for all classes
and depicts the raw frequencies/counts in each class. By dividing the raw
counts by n times the class width the density histogram is found where
the area of all bars sum to 1 26–28, 31

(Statistical) Inference [Statistisk inferens (følgeslutninger baseret på data)] 5

Inter Quartile Range [Interkvartil bredde] The Inter Quartile Range (IQR) is
the middle 50% range of data 15

Linear regression [Lineær regression (-sanalyse)] 1, 20

Median [Median, stikprøvemedian] The median of population or sample (note,

in text no distinguishment between population median and sample median)
7, 9, 10, 23

Multiple linear regression [Multipel lineær regression (-sanalyse)] 1

Quantile [Fraktil, stikprøvefraktil] The quantiles of population or sample (note,

in text no distinguishment between population quantile and sample quantile)
11

Quartile [Fraktil, stikprøvefraktil] The quartiles of population or sample (note,

in text no distinguishment between population quartile and sample quartile)
11

Sample variance [Empirisk varians, stikprøvevarians] 13

Sample mean [Stikprøvegennemsnit] The average of a sample 8, 10, 13, 23

Chapter 1 Acronyms 44

Acronyms

ANOVA Analysis of Variance Glossary: Analysis of Variance

cdf cumulated distribution function Glossary: cumulated distribution function

CI confidence interval Glossary: confidence interval

CLT Central Limit Theorem Glossary: Central Limit Theorem

IQR Inter Quartile Range 8, 15, 29, 30, Glossary: Inter Quartile Range

LSD Least Significant Difference Glossary: Least Significant Difference

pdf probability density function Glossary: probability density function