APPLIED STATISTICS

TEXTBOOK

Doc. RNDr. ZdenČk Karpíšek, CSc.

Ing. Miloš Drdla, Dr., MBA

BRNO 2007

(Second Edition)
Doc. RNDr. ZdenČk Karpíšek, CSc.
Department of Statistics and Optimization
Institute of Mathematics
Faculty of Mechanical Engineering
Brno University of Technology
Technická 2, 616 69 Brno
Phone: +420 541 142 529
E-mail: [email protected]

Ing. Miloš Drdla, Dr., MBA

Brno International Business School
Lidická 81, 602 00 Brno
Phone: +420 545 570 111
E-mail: [email protected]

© ZdenČk Karpíšek 2007

Miloš Drdla

2
CONTENT

PREFACE (5)

1. FUNDAMENTALS OF PROBABILITY (7)

Random events (7)
Axiomatic definition of probability and its properties (9)
Conditioned probability and independent events (11)
Random variable and its probability distributions (14)
Numerical characteristics of random variables (18)
Random vectors and their probability distributions (20)
Numerical characteristics of random vectors (22)
Some important probability distributions (25)

2. DESCRIPTIVE STATISTICS (33)

Basic notions (33)
Processing univariate sample data with a quantitative variable (34)
Processing bivariate sample data with quantitative variables (44)
Processing sample data with qualitative variables (48)
Exercises (48)
Questions (52)

3. ANALYSIS OF TIME SERIES (53)

Fundamentals (53)
Interval and point time series (55)
Special types of time series (58)
Development of time series (61)
Time series analysis (65)
Exercises (69)
Questions (71)

3
4. INDEX NUMBERS (72)
Basic notions (72)
Simple index numbers (72)
Group indexes (73)
Composite indexes (75)
Index numbers and absolute quantities (78)
Basic and chain indexes (79)
Exercises (80)
Questions (82)

5. MATHEMATICAL STATISTICS (83)

Random sample and its characteristics (83)
Parameter estimation (84)
Testing statistical hypotheses (90)
Regression analysis (100)
Exercises (107)
Questions (116)

BIBLIOGRAPHY (118)

STATISTICAL TABLES (119)

4
 PREFACE

Modern times have created modern problems and many of those problems
involve data. Marketing studies, product testing and quality control are typical
application areas that require an intelligent analysis of data. In particular, the
business and manufacture now demand employers as well as employees that are
better prepared to use statistics.

This textbook is introduction into statistics and consists of five chapters,

statistical tables, and bibliography.

Chapter No1: Fundamentals of Probability

Probability plays a special role in all our lives, because we use it to measure
uncertainty. We are continually faced with decisions leading to uncertain outcomes
and we rely on probability to help us make our choice. A probability is a numerical
value that measures the uncertainty that a particular event will occur. The probability
of an event ordinarily represents the proportion of times under identical
circumstances that the event can be expected to occur.

Chapter No2: Descriptive Statistics

The emphasis on the decision-making aspects of statistics is a recent one. In

its early years, the study of statistics consisted of methodology for summarizing or
describing numerical data. This area of study has become known as descriptive
statistics because it is concerned with summary calculations and graphical displays.
These methods are in contrast to the modern statistical approach in which
generalizations are made about the whole (called the population) by investigating a
portion (referred to as the sample).

Chapter No3: Analysis of Time Series

When data for evaluations are collected at regular intervals from monthly,
quarterly, or annual reports, they are referred to as time-series data. In each case,
values of the variable being predicted are available for several past periods of time.
Such data are called time-series. Statistical procedures that use such values are
called time-series analysis.

5
 Chapter No4: Index Numbers

Index numbers are used to facilitate comparisons of time-series data from

different periods. Price indexes are especially useful in expressing price changes
over time, thereby gauging inflation, a condition of rising prices that permeates the
modern world.

Chapter No5: Mathematical Statistics

Mathematical statistics is a body of methods and theory that it applied to

numerical evidence when making decisions in the face of uncertainty. This
specification treats mathematical statistics as a relatively independent academic
field. Using the methods of mathematical statistics, we can describe quantities of
random character from observed values. Most frequently, we try to establish the
properties of the probability distribution of a random variable - estimating parameters
or quantitative characteristics, testing the hypotheses that they have certain
properties, analyzing the relationships between them etc.

In writing this introductory statistics textbook for students of business and

management, our overriding goal has been to enliven statistics, to make it more
interesting and relevant and easier to learn. To illustrate that statistics is neither
boring nor irrelevant this textbook treats it as essentially a decision-making tool and
includes some relevant concepts and possible applications. We are greatly indebted
to RNDr. Karel Mikulášek Ph.D. who has assisted us in preparing and translating this
textbook.

Brno Doc. RNDr. ZdenČk Karpíšek, CSc.

October 2007 Ing. Miloš Drdla, Dr., MBA

6
 1. FUNDAMENTALS OF PROBABILITY

Random events

An experiment is any well-defined situation or procedure that results in one of

two or more possible outcomes. An outcome is a particular result of an experiment. A
random experiment is an experiment where there is uncertainty about which of the
possible outcomes will occur. We will assume that a random experiment may be
repeated a sufficiently large number of times while the conditions that define the
random experiment remain constant.

An elementary event {Z} expresses individual outcomes in their simplest

terms, that is, in the most elementary form. The set : of all possible outcomes for an
experiment is called the sample space. A random event (or event) A is a subset of :.
If {Z}A, and {Z} is the outcome of a random experiment, we say that a random
event A has occurred as a result of the experiment. We will denote random events by
A, B, A1, Ai, … . An event that unavoidably occurs for every random experiment is
called certain (sure). It is equivalent to :. An event that cannot occur for any random
experiment is called impossible and is equivalent to the empty set ‡.

Relationships between random events are expressed in terms of set

inclusions:

a) A Ž B means that the occurrence of random event A results in the occurrence

of random event B.

b) A = B expresses the equality of random events A and B.

Operations with random events are expressed in terms of set operations:

a) The union A ‰ B of random events A and B occurs if at least one of the

n
random events A and B, that is, A and/or B occurs. In a similar way, we define *A
i 1
i

f
and *A
i 1
i .

7
 b) The intersection A ˆ B of random events A and B occurs when both events
n f
occur. Similarly, we define A
i 1
i and A
i 1
i .

c) The difference A – B of random events A and B occurs when A occurs and B

does not occur.

d) The complementary event to random event A is the event A :A.

e) Random events A and B are called exclusive if A ˆ B = ‡.

A field of events 6 is a set of random events (a system of subsets of the

sample space :) with the following properties:

‡, : 6,

A6 Ÿ A6,
f
Ai 6, i = 1, 2, … Ÿ A 6 .
i 1
i

E x a m p l e 2.1

A random experiment is made by throwing a dice from homogeneous material with

faces numbered from 1 to 6. Event A occurs when an even number comes up and
event B occurs if a number greater than 4 comes up. Determine :, A, B , A ‰ B,

A ˆ B, A  B, B  A, 6.

S o l u t i o n:

The sample space : = {1, 2, 3, 4, 5, 6} with elementary events {1}, {2}, {3}, {4}, {5},
{6}. Next we have A = {2, 4, 6} a B = {5, 6} so that

A = {1, 3, 5} ... an odd number comes up,

B = {1, 2, 3, 4} ... a number less than 5 comes up,

A ‰ B = {2, 4, 6} ‰ {5, 6} = {2, 4, 5, 6} ... numbers 1 a 3 do not come up,

A ˆ B = {2, 4, 6} ˆ {5, 6} = {6} ... number 6 comes up,

A  B = {2, 4, 6}  {5, 6} ={2, 4} ... number 2 or 4 comes up,

8
 B  A = {5, 6}  {2, 4, 6} = {5} ... number 5 comes up.

Since no restrictions are imposed on random events, we can consider the maximal
field of events (the system of all subsets of :):

6 = {‡, {1}, {2}, {3}, {4}, {5}, {6}, {1,2}, {1,3},…,{5,6},…, {2, 3, 4, 5, 6}, : }
which contains 26 = 64 random events.

Axiomatic definition of probability and its properties

The probability P(A) of a random event A6 is a non-negative function

(defined on 6) such that P(:) = 1 and, for each sequence of mutually exclusive
§f · f
random events Ai  6, i = 1, 2, …, we have P¨¨ * A i ¸¸ ¦P A i .
©i 1 ¹ i 1

It holds:

a) P( A) 1  P A ; P(‡) = 0; 0 d P(A) d 1.

b) A Ž B Ÿ P(A) d P(B) and P(B – A) = P(B) – P(A).

c) P A 1 ‰ ... ‰ A n 1  P( A 1 ˆ ... ˆ A n )
n n
n1
¦P A i  ¦P A i ˆ A j  ...   1
i 1 i, j 1
P A 1 ˆ ... ˆ A n .
i¢ j

This means particularly that P A ‰ B 1  P( A ˆ B ) P A P B P A ˆB .

For a finite or countable sample space : (i.e. its elementary events ^Z` may
be listed in a sequence) we get
PA ¦ P ^Z` .
Z A

Particularly, for a sample space : with n equally probable elementary events we

have
m
P( A ) ,
n
where m is the number of elementary events ^Z` of which random event A consists.
We say that "m is the number of favourable outcomes of the experiment" and "n is

9
the number of outcomes of the experiment". This is the so-called classical definition
of probability.

E x a m p l e 2.2

Calculate the probabilities P(A), P(B), P( A ), P( B ), P(A ‰ B), P(A ˆ B), P(A  B),
P(B  A) of the random events from Example 2.1.

S o l u t i o n:

Due to the symmetry and homogeneity of the cube, all the elementary events have
the same probability P({Z}) = 1/6 and n = 6. This yields the following probabilities:

P(A) = 3/6 = 1/2,

P(B) = 2/6 = 1/3,

P( A ) = 3/6 = 1/2,

P( B ) = 4/6 = 2/3,

P(A ‰ B) =4/6 = 2/3,

P(A ˆ B) = 1/6,

P(A  B) = 2/6 = 1/3,

P(B  A) = 1/6.

Using the properties of probability we can, for example, calculate

1 1
P( A ) = 1 P(A) = 1  ,
2 2

1 1 1 2
P(A ‰ B) = P(A) + P(B)  P(A ˆ B) =   .
2 3 6 3

E x a m p l e 2.3

In a supply of 100 shafts, 10 items do not comply with the standard diameter,
20 items have not the required length and 5 items comply neither with the length nor
with the diameter requirement. Calculate the probability that a shaft selected at
random has both the required length and diameter.

10
S o l u t i o n:

Denoting by A and B the event that the selected shaft does not comply with the
required diameter and length respectively, the probability that the selected shaft has
both the required length and diameter

P( A ˆ B ) 1  P A ˆ B 1  P( A ‰ B) 1  >P( A)  P(B)  P( A ˆ B)@

1  (0.10  0.20  0.05) 0.75 .

Conditioned probability and independent events

The probability of event A  6, given event B  6, P(B) z 0, is called the

conditional probability of event A

P( A ˆ B)
P( A / B) .
P(B)

It holds:

a) P(A1 ˆ … ˆ An) = P(A1)P(A2/A1) … P(An/A1 ˆ … ˆ An – 1),

particularly P(A ˆ B) = P(A)P(B/A) = P(B)P(A/B),

n
b) For a random event A Ž * B i where Bi are mutually exclusive random events,
i 1

i = 1, …, n, the probability
n
P( A ) ¦ P(B )P( A / B )
i 1
i i

is calculated by the so-called formula of total probability and, for P(A) z 0, we

can derive Bayes' formulas for the probability of hypotheses or Bayes'
theorem

P(B j )P( A / B j )
P(B j / A ) n
, j = 1, …, n.
¦ P(B )P( A / B )
i 1
i i

11
E x a m p l e 2.4

Ten products out of a total of 100 are defective. We choose 3 products at random
without replacement. The probability that the first product chosen is defective -
random event A1, the second product chosen is defective - random event A2, and the
third product chosen is not defective – random event A 3 , is calculated below:

P( A1 ˆ A 2 ˆ A 3 ) P( A1 )P( A 2 / A1 )P( A 3 / A1 ˆ A 2 )
90 / 100 89 / 99 10 / 98 # 0.08256 .

E x a m p l e 2.5

A grocery receives supplies of croissants from 3 bakehouses in the following

quantities: 500, 1000, and 1500 croissants daily. The percentages of defective
croissants for the suppliers are 5%, 4%, and 3%. In the grocery, the croissants from
individual suppliers are mixed together and sold. A croissant is bought at random.
Calculate the probability that

a) it is defective,

b) it has been supplied by the second bakehouse.

S o l u t i o n:

Denote by A the random event that the croissant bought is defective and by Bi,
i = 1, 2, 3, the event that the croissant has been supplied by the i-th bakehouse. We
get the following probabilities

500 1
P(B1 ) , P( A / B1 ) 0.05 ,
500  1000  1500 6

1000 2
P(B 2 ) , P( A / B 2 ) 0.04 ,
500  1000  1500 6

1500 3
P(B 3 ) , P( A / B3 ) 0.03 .
500  1000  1500 6

Using the formula of total probability we calculate

1 2 3 0.22
P( A) (0.05)  (0.04)  (0.03) 0.036 # 0.03667 ,
6 6 6 6

12
so that, from the customer's point of view, the chance of buying a defective croissant
is approximately 3.667%. Applying Bayes' theorem for j = 2, we have

2
(0.04)
6 0.08
P(B 2 / A) 0.36 # 0.36364 .
0.22 0.22
6

In a similar way, we can obtain P(B1 /A) # 0.22727 and P(B3 /A) # 0.40909, which
means that the third bakehouse supplies the largest quantity even if their percentage
of defective products is the lowest of the three. This is due to the fact that they supply
the largest part of the croissants.

Random events A, B  6 are called independent, if P(A/B) = P(A) or P(B) = 0.

Random events A1, …, An  6 are mutually independent, if all the pairs of random
events
Ai, Aj for i z j,

Ai, Aj ˆ Ak for i z j, i z k,

Ai, Aj ˆ Ak ˆ Am for i z j, i z k and i z m,

etc.
are independent.

The following assertions are true:

a) A, B are independent if and only if P(A ˆ B) = P(A)P(B).

b) If A1,…, An are mutually independent, then

x P(A1 ˆ … ˆ An) = P(A1) … P(An),

x P(A1 ‰ … ‰ An) = 1 - >1 – P(A1)@ … >1 – P(An)@,

x B1,…, Bn are mutually independent for arbitrary variants B i A i, A i, : .

E x a m p l e 2.6

What is the probability that, when throwing a dice, an even number comes up in the
first trial, (random event A) and, in the second trial, an odd number comes up
(random event B)?

13
S o l u t i o n:
Random events A and B are independent and their probabilities are P(A) = P(B) =
= 1/2, so that P(A ˆ B) = (1/2)(1/2) = 1/4.

E x a m p l e 2.7

A product is processed by three independent operations, where the chances of

producing a defective product are P(A1) = 0.05, P(A2) = 0.08 and P(A3) = 0.03.
Calculate the probability that, after having been processed by all the three
operations, the product is defective.

S o l u t i o n:

Since the operations are independent, the random events A1, A2, A3 are mutually
independent and the product is defective if at least one of them occurs so that

P(A1 ‰ A2 ‰ A3) = 1  >1 – P(A1)@>1 – P(A2)@>1 – P(A3)@ = 1 – (0.95)(0.92)(0.97) =

= 0.152220.

Random variable and its probability distributions

A random variable X is a variable that takes on real numbers x and whose

outcomes occur by chance – for details see [1], [2], [3] and [4]. Its distribution function
of probabilities (or just distribution function) is given by

F(x) = P(X  x) = P>X(-f; x)@, x(-f;+f).

The distribution function has the following properties:

a) 0 d F(x) d 1 for all x(-f;+f),

b) F(x) is non-decreasing and continuous on the left in (-f;+f).

c) lim F( x ) 0 , lim F( x ) 1 ,
x o f x o f

d) P(a d X  b) = F(b) – F(a) for arbitrary real numbers a  b,

e) P( X c) lim F( x)  F(c) for any real c.

xo c 

14
 A random variable X is said to be discrete with a discrete distribution of
probabilities if it takes on at most a countable number of values x = x1, x2,… . Its
probability distribution is given in the form of a sequence

p(x) = P(X = x) ! 0 for x = x1, x2, ….

We have:
a) ¦ p(x)
x
1,

b) F( x ) ¦ p(t )
t¢x
for all x(-f;+f),

c) P X M ¦ p( x) for an arbitrary set of real numbers M.

xM

The distribution function for a discrete random variable is a "step-like line with jumps"
– see Fig. 2.1.

Prob. Mass Fcn. Cum. Dist. Fcn. n = 3, p = 0.5

n = 3, p = 0.5
Binomial Binomial

0.5 1

0.4 0.8

0.3 0.6
prob. mass cum. prob.

0.2 0.4

0.1 0.2

0 0

0 1 2 3 -1 0 1 2 3 4
X X

(a) (b)

Fig. 2.1 The graphs of the distribution of probabilities (a) and the distribution function
(b) of a discrete variable.

E x a m p l e 2.8

The probability of failure for each of three independently operating production lines is
0  p  1. The discrete random variable X that expresses the number of production
lines with a failure takes on the values x = 0, 1, 2, and 3 and the values of its
distribution of probabilities are given below

15
 p(0) = (1 – p)3,

p(1) = 3p(1 – p)2,

p(2) = 3p2(1 – p),

p(3) = p3.

Its distribution function is the following

F(x) = 0 for x  (-f, 0²,

F(x) = p(0) = (1 – p)3 for x  (0, 1²,

F(x) = p(0) + p(1) = (1 + 2p)(1 – p)2 for x  (1, 2²,

F(x) = p(0) + p(1) + p(2) = (1 + p + p2)(1 – p) = 1 – p3 for x  (2, 3²,

F(x) = p(0) + p(1) + p(2) + p(3) = 1 for x  (3; f).

In Fig. 2.1 you can see the graphs of p(x) and F(x) for p = 0.5. The probability of a
failure occurring in at least one of the production lines is given by

P(X t 1) = P(1 d X  +f) = F(+f)  F(1) = 1 – (1 – p)3.

A random variable X is said to be continuous with a continuous distribution of

probabilities if its distribution function is continuous (thus X takes on all the values of
an interval, etc.). Its density function, is such a non-negative function f(x) that
x

F( x) ³ f ( t)dt
f
for all x(-f;+f).

It has the following properties:

f

a) ³ f ( x)dx
f
1,

b) f(x) = Fc( x ) , if the derivative exists,

c) F(x) is a continuous function for all x(-f;+f),

b
d) P(a d X d b) P(a  X  b) P(a  X d b) P(a d X  b) ³ f (x)dx
a
F(b)  F(a)

for arbitrary real numbers a d b,

e) P(X = c) = 0 for any real c.

16
E x a m p l e 2.9

A real variable X has a density function f(x) = cx for x  ¢0; 2² and 0 for x  ¢0; 2².
Using the properties of a continuous random variable, we can derive the following
results. We have
f 0 2 f

³ f ( x )dx ³ 0dx  ³ cxdx  ³ 0dx

f f 0 2
... 2c 1,

which yields c = 1/2. The distribution function of X is given by

x
Fx ³ 0dt
f
0 for x  (- f; 0²,

0 x
t x2
F( x ) ³f0dt  ³0 2dt ... 4
for x  ¢0; 2²,

0 2 x
t
F( x ) ³f0dt  ³0 2dt  ³2 0dt ... 1 for x  ¢2;+f).

The graphs of f(x) and F(x) are shown in Fig. 2.2. The probability of the random
variable taking on a value x  ¢1; 3² is P(1 d X d 3) = F(3) – F(1) = 1 – (12/4) = 0,75.

1.5 1.5

1 1

f(x) F(x)

0.5 0.5

0 0

-2 -1 0 1 2 3 4 -2 -1 0 1 2 3 4
x x

(a) (b)

Fig. 2.2 Graphs of the density function (a) and the distribution function (b) of a
continuous random variable.

17
 Numerical characteristics of random variables

Numerical characteristics of random variables are real numbers that, in a

concentrated way, express their important properties.

The central tendency is characterized by the expected value or mathematical

expectation of a random variable X

E( X) ¦ xp( x )
x
for discrete random variable X,

f
E( X) ³ xf ( x)dx
f
for continuous random variable X,

provided that the sum or the integral is absolutely convergent. The expected value
has the following properties:

a) E(aX + b) = aE(X) + b for arbitrary real numbers a, b,

§ n · n
b) E¨¨ ¦ X i ¸¸ ¦E X i for random variables X1,…,Xn.
©i1 ¹ i 1

A measure of the variation of a random variable X about its expected value

E(X) is expressed by its variance (dispersion)

>
D( X) E ( X  E( X)) 2 . @
The variance has the following properties:

a) D( X) ¦ ( x  E( X))
x
2
p( x ) ¦ x p( x)  (E( X))
x
2 2
for discrete random variable X,

f f
2 2
b) D( X ) ³ (x  E( X)) f ( x )dx ³x f ( x )dx  (E( X ))2 for continuous random
f f

variable X, provided that the sum or the integral converges.

c) D(X) t 0,

d) D(aX + b) = a2 D(X) for arbitrary real numbers a, b,

§ n · n
e) D¨ ¦ X i ¸ ¦D X i for independent random variables X1,…, Xn.
©i1 ¹ i 1

18
 The standard deviation of a random variable X is defined as V( X ) DX .

The standard deviation has the following properties:

a) V(X) t 0;
b) V(aX + b) = _a_ V(X) for arbitrary real numbers a, b.

The mathematical expectation, is one of the so-called moments about the

origin while the variance is a special case of the so-called central moment. You can
learn more about moment characteristics such as the coefficient of variation,
coefficient of skewness, coefficient of pointedness in [1], [2], [3] and [4].

For a 0  P  1, the P-quantile or the 100P%-quantile (percentile) of a random

variable is defined as xP = max ^x; F(x) d P`. For a continuous random variable X
with an increasing distribution function we have F(xP) = P. The median of a random
variable X is its quantile x0,5 and it is characteristic of its position. Further quantile
characteristics can be found in >2@, >3@. The mode x of a random variable X is its
value for which the probability distribution or the density function assumes a
maximum value or a supremum.

E x a m p l e 2.10

The random variable X from Example 2.9 has the expected value
0 2 f
x 4
EX ³ x ˜ 0dx  ³ x
f 0
2
dx  ³ x ˜ 0dx
2
...
3
# 1.33333 ,

the variance
0 2 f 2
2 x 2 2 §4· 16 2
D( X ) ³fx ˜ 0dx  ³0 x 2 dx  ³2 x ˜ 0dx  ¨© 3 ¸¹ 2
9 9
# 0.22222 ,

and the standard deviation

2
VX # 0.47140 .
9
x2
The P-quantile xP is the root of the equation P that lies in the interval ¢0; 2²,
4
which means that xP = 2 P so that the median of X is x0,5 = 2 0.5 # 1.41421. From
the graph of f(x) in Fig. 2.2, we can see that the mode of X is x = 2.

19
 Random vectors and their probability distributions

A two-dimensional random vector (a bivariate random variable) is an ordered

pair of random variables (X,Y). In [1], [2], and [3], you can learn about n-dimensional
random vectors in general . The joint distribution function of a random vector (X,Y) is
defined as

F(x,y) = P(X x, Y y) = P>(X,Y)(-f; x)u (-f; y)@, (x,y)(-f;+f)2.

It has the following properties:

a) 0 d F(x,y) d 1 for all pairs (x,y) (-f;+f)2,

b) lim F( x, y) F( f, y) lim F( x, y) F( x,f) 0,

x o f y o f

c) lim F( x, y) F( f,f) 1,
( x ,y )o  f,f

d) F(x,y) is non-decreasing and continuous on the left for each variable x

and y.

A random vector (X,Y) is said to be discrete with a discrete joint distribution

function, if both its constituents X and Y are discrete and therefore it assumes at
most a countable number of values (x,y) = (x1,y1), (x2,y2),… . The joint distribution of
probabilities of a discrete random vector is the sequence p(x,y) = P(X = x,Y = y) ! 0.
We have:

a) ¦¦ p(x, y)
x y
1,

b) F( x, y) ¦¦ p(u, v) for all pairs (x,y) (-f;+f)2,

u x v  y

c) P ( X, Y)  M ¦¦ p(x, y) for M Ž (-f;+f)2.

( x ,y )M

A random vector (X,Y) is said to be continuous with a continuous joint

distribution function if both its constituents X, Y are continuous. The joint density
function of a continuous random vector is such a function f(x,y) that
x y
F( x, y) ³ ³ f (u, v)dudv for all pairs (x,y) (-f;+f)2.
 f f

20
We have:
f f
a) ³ ³ f (x, y)dxdy 1,
 f f

w 2F( x, y )
b) f ( x, y ) provided that the partial derivative exists,
wxwy

c) P ( X, Y )  M ³ ³ f ( x, y)dxdy for M Ž (-f;+f)2.

M

If, in a random vector (X,Y), we leave out its constituent X or Y (that is, we
assume that this random variable takes on any value), we get marginal probability
distributions for Y or X. For marginal distributions of probability, density functions,
and distribution functions we have

a) p X (x) ¦ p( x, y ),
y
p Y (y) ¦ p( x, y)
x
for a discrete random vector,

f f
b) fX ( x) ³ f ( x, y)dy,
f
fY (y) ³ f ( x, y)dx
f
for a continuous random vector,

c) FX ( x ) F( x,f ), FY ( y ) F( f, y ) for both a discrete and continuous

random vector.

Further we use conditioned probability distributions for one random variable

given that the other random variable takes on an arbitrary but fixed value. In this way
we get the conditioned probability distribution of a random variable X or Y given that
Y = y or X = x respectively. For conditioned distributions of probability pX(x/y) =
= P(X = x / Y = y) and the like, and conditioned density functions we have

p( x, y ) p( x, y )
a) pX (x / y) , pY (y / x) for a discrete random vector,
pY (y) p X (x)

f ( x, y ) f ( x, y )
b) fX ( x / y) , fY (y / x) for a continuous random vector.
fY (y) f X (x)

Random variables X and Y are said to be independent if, for all pairs (x,y),

p(x,y) = pX(x)pY(y) or f(x,y) = fX(x)fY(y), and F(x,y) = FX(x)FY(y).

21
 Numerical characteristics of random vectors

A numerical characteristic of a random vector expresses its important

properties. The central tendency of a joint probability distribution is characterized by
the centre (E(X), E(Y)) of a random vector (X,Y) where E(X) and E(Y) are the
expected values of X and Y. We can write

a) E( X) ¦ xp
x
X ( x) ¦¦ xp( x, y ) ,
x y
E( Y ) ¦ yp
y
Y (y) ¦¦ yp( x, y)
x y
for a

discrete random vector,

f f f f f f
b) E( X) ³ xf
f
X ( x )dx ³ ³ xf ( x, y)dxdy , E( Y ) ³ yf
 f f f
Y ( y )dy ³ ³ yf ( x, y )dxdy
 f f
for

a continuous random vector, provided that the sums or integrals

absolutely converge.

The relationship between the random variables X and Y of a random vector

(X,Y) is given by their covariance

cov( X, Y ) E>( x  E( X))( y  E( Y ))@ .

The covariance has the following property:

a) cov( X, Y ) ¦¦ ( x  E( X))( y  E( Y ))p( x, y) ¦¦ xyp( x, y )  E( X)E( Y )

x y x y
for

a discrete random vector,

f f f f
cov( X, Y ) ³ ³ ( x  E( X))( y  E( Y ))f ( x, y)dxdy ³ ³ xyf( x, y )dxdy  E( X)E( Y )
 f f  f f

for a continuous random vector, provided that the sums or integrals

absolutely converge.

b) cov(X,Y) = cov(Y,X),

c) cov(X,X) = D(X),

d) D(X + Y) = D(X) + D(Y) + 2cov(X,Y),

e) X, Y independent Ÿ cov (X,Y) = 0 and E(X,Y) = E(X)E(Y).

22
 The covariances of a random vector (X,Y) can be displayed in the form of a
symmetrical covariance matrix cov(X,Y) – see [1], [2], [3] and [4].

A measure of the linear independence between the random variables X and Y

is their correlation coefficient (coefficient of correlation)

cov X, Y
U X, Y .
D XDY

The correlation coefficient has the following properties:

a) U(X,Y) = U(Y,X),

b) U(X,X) = U(Y,Y) = 1,

c) -1 d U(X,Y) d 1,

ac
d) U(aX + b, cY + d) = U( X, Y ) for arbitrary real numbers a, b, c, d, ac z 0,
ac

e) Y = aX + b œ _U(X, Y)_ = 1 where a, b are real numbers, a z 0,

f) X, Y independent Ÿ U(X, Y) = 0.

If U(X,Y) = 0, we say that the random variables X and Y are not correlated.

Independent random variables are not correlated but two random variables
that are not correlated are not necessarily independent. However, if they are both
normally distributed, they are also independent. The correlation coefficients of a
random vector (X,Y) can be displayed in the form of a symmetrical correlation matrix
U(X,Y) - see [1], [2], [3] and [4].

E x a m p l e 2.11

A discrete random vector (X,Y) is given by the below table:

23
 x

y 0 1 2 3

-1 2c c 0 0

0 c 2c c 0

1 0 0 2c c

Calculate c, F(2;0), pX(1), FY(1), pX(x/y) for (x,y) = (1;0), E(X), E(Y), D(X), D(Y),
cov(X,Y), U(X,Y) and determine if the random variables X,Y are independent.

S o l u t i o n:

2c + c + ... + 2c + c = 1 Ÿ 10c = 1 Ÿ c = 0.1;

F(2;0) = p(0;-1) + p(1;-1) = 0.3 ;

pX(1) = p(1;-1) + p(1;0) + p(1;1) = 0.1 + 0.2 + 0 = 0.3 ;

FY(1) = pY(-1) + pY(0) = 0.3 + 0.4 = 0.7 ;

pX(1/0) = p(1;0)/pY(0) = 0.2/0.4 = 0.5 ;

E(X) = (0)(0.2) + (0)(0.1) + ... + (3)(0) + (3)(0.1) = 1.2 ;

E(Y) = (-1)(0.2) + (-1)(0.1) + ... + (1)(0.2) + (1)(0.1) = 0 ;

D(X) = (02)(0.2) + (02)(0.1) + ... + (32)(0) + (32)(0.1) – 1.22 = 2.4 - 1.44 = 0.96 ;

D(Y) = (-1) 2(0.2) + (-1) 2(0.1) + ... + 12(0.2) + 12(0.1) - 02 = 0.6 - 0 = 0.6 ;

cov (X,Y) = (0)(-1)(0.2) + (1)(-1)(0.1) +...+ (2)(1)(0.2) + (3)(1)(0.1) – (1.2)(0) =

= 0.6 - 0 = 0.6 ;

0.6
U(X,Y) = # 0.79057 z 0, which means that X, Y are not
(0.96)(0.6)

independent.

24
 Some important probability distributions

Discrete probability distributions

a) The binomial probability distribution Bi(n, p) where n is a natural number

and p is a real number, 0  p  1:

§ n· x n x
p( x) ¨ ¸ p 1 p , x = 0, 1, …, n;
© x¹

E(X) = np; D(X) = np(1 – p); (n + 1)p – 1 d x d (n + 1)p.

This is the probability distribution of a random variable that expresses the number of
occurrences of an observed event in the sequence of n mutually independent trials
(such as the number x of defective products out of a total of n products if p is the
probability of a defective product being manufactured). This probability distribution
may be employed when performing a random sample with replacement such as
checking n products from a supply and replacing each product after it has been
checked. For np(1 – p) ! 9 the binomial distribution may be approximated by a
normal distribution where P = np, V2 = np(1 – p). For p  0,1 and n ! 30 we can also
approximate this probability distribution by the Poisson probability distribution with
O = np. The graphs of the distribution of probability and the distribution function of the
binomial distribution for n = 3 and p = 0,5 are shown in Fig. 2.1.

E x a m p l e 2.12

A series of 50 products contains 5 defective ones. Three products are selected at

random from the series. The number of defective products among those selected is a
random variable X. Determine its type of probability distribution, its distribution of
probabilities p(x), expected value E(X), variance D(X), standard deviation V(X),
median x0.5, mode x , and the probability P(1 < X d 3). Each product selected is
replaced so that this is the so-called random sample with replacement.

S o l u t i o n:

The random variable X has the binomial distribution Bi(n,p) with n = 3 and p = 5/50 =
= 0.1. X takes on the values x = 0, 1, 2, 3. The distribution of probabilities

25
 § 3 · x 3 x
p( x ) ¨¨ ¸¸0,1 0,9 for x = 0, 1, 2, 3.
©x¹

The expected value E(X) = np = (3)(0.1) = 0.3,

the variance D(X) = np(1 - p) = (3)(0.1)(0.9) = 0.27,

the standard deviation V(X) = D(X ) 0.27 # 0.51962,

the median x0.5 = 0, since p(0) = 0.729,

the mode x = 0, since (n + 1)p -1 = -0.6 and (n + 1)p = 0.4,

P(1 < X d 3) = p(2) + p(3) = 0.027 + 0.001 = 0.028.

b) The hypergeometric distribution H(N,M,n) where N, M, and n are natural

numbers, 1 d n d N, 1 d M d N:

§ M· § N  M·
¨ ¸¨ ¸
© x ¹ ©n  x ¹
p( x) , x = max ^0, M – N + n`, …, min ^M, N`;
§ N·
¨ ¸
©n¹

M M § M· N  n M1 n 1
EX n ; DX n ¨1  ¸ ; a – 1 d x d a where a .
N N © N¹ N  1 N 2

This probability distribution describes the so-called random sample without

replacement where, for example, N is the total number of products to be checked, M
is the number of defective products among them, n products are selected at random
without replacement, and x is the number of defective products among those
selected. For n/N  0,1, the hypergeometric probability distribution may be
approximated by the binomial probability distribution with p = M/N, or, for n/N  0,1,
M/N  0,1 and n ! 30, by the Poisson probability distribution with O = nM/N.

E x a m p l e 2.13

Among a total of 50 products there are 5 defective ones. Three products are drawn at
random out of the total. The number of defective products among those selected is a
random variable X. Determine the type of its probability distribution, its distribution of

26
probabilities p(x), expected value E(X), variance D(X), standard deviation V(X),
median x0,5, mode x , and P(1 < X d 3). Assume, as opposed to Example 2.12, that a
product is not replaced once it has been selected so that this is a random sample
without replacement.

S o l u t i o n:

The random selection X has the probability distribution H(N,M,n) with N = 50, M = 5
and n = 3. X takes on the values x = 0, 1, 2, and 3. The distribution of probabilities is
given by
§ 5 ·§ 45 ·
¨¨ ¸¸¨¨ ¸¸
p( x ) © x ¹© 3  x ¹ for x = 0, 1, 2, 3.
§ 50 ·
¨¨ ¸¸
©3¹

M
The expected value E(X) = n = (3)(0.1) = 0.3,
N
M § M· N  n
the variance D(X) = D X n ¨1  ¸ = (3)(0.1)(0.9) (47/49) | 0.25898,
N © N¹ N  1

the standard deviation V(X) = D(X ) 0.25898 # 0.50890,

the median x0.5 = 0 since max p(x) = p(0) # 0.724,

M1 n 1
the mode x = 0 since a # 0.46154, a – 1 # -0.53846,
N 2

P(1 < X d 3) = p(2) + p(3) # 0.023 + 0.0005 = 0.0235.

c) The Poisson probability distribution Po(O) where O is a real number, O ! 0:

Ox  O
p( x) e , x = 0, 1, … ; E(X) = O; D(X) = O; O - 1 d x d O.
x!

This probability distribution is usually used to determine the probability of a number of

occurrences of an observed event within a time interval (number of failures,
accidents, disasters, defective products and the like) with a small probability of
occurrence.

27
E x a m p l e 2.14

On the average, three customers enter a shop within a given minute. Determine the
appropriate type of probability distribution of a random variable that expresses the
number of customers that enter the shop within a given minute, the expected number
of customers, the variance of this number, and the most likely number of customers
that enter the shop within a given minute. Next calculate the probability that within
that minute a) exactly one customer enters the shop, b) at least one customer enters
the shop.

S o l u t i o n:

If we approximate the expected value of customers that enter the shop within one
given minute by their average number, we can assume that the random variable X
has the Poisson probability distribution Po(O) with the distribution of probabilities
given by

3x 3
p( x ) e , x = 0, 1, …
x!

The expected value E(X) = O = 3,

The variance D(X) = O = 3,

for the mode we have O - 1 d x d O, which yields x = 2 a 3,

31 3
P(X = 1) = p(1) = e # 0.14936,
1!

30  3
P(X t 1) = p(1) + p(2) + … = 1 – p(0) = 1 – e # 1 – 0.04979 = 0.95021.
0!

Information on further numerical characteristics of the above discrete

probability distributions and multi-dimensional probability distributions can be found
in [1], [2], [3] and [4].

28
 Continuous probability distributions

a) The uniform probability distribution R(a, b) where a  b are real numbers:

1
f (x ) for x  a;b ,
ba
0 for x  a;b ,

F( x ) 0 for x   f; a ,
xa
for x  ¢ a; b²,
ba
1 for x  b;f ,

ab (b  a) 2
EX x 0.5  D( X ) .
2 12

The graphs of the density function and the distribution function for a = -1 and b = 2
are shown in Fig. 2.3. This probability distribution is mostly used to simulate real
processes, in numerical calculations to implement the so-called Monte Carlo method
on a computer, and for calculations using the so-called geometric probability.

E x a m p l e 2.15

An optical cable of a length of 500 m may be disrupted at any distance from its
beginning. The probability of the random event that the cable will be disrupted in a
given section is in direct proportion to the length of the section and is independent of
its position. Determine the probability distribution of the random event X expressing
the distance of a disruption from the beginning of the cable, the density function, the
basic numerical characteristics, and the probability that the cable will be disrupted in
the section beginning at 300 m and ending at 400 m.

S o l u t i o n:

Random variable X has the probability distribution R(a, b) with a = 0 and b = 500.

The density function is given by

1
f (x ) for x  0; 500 ,
500
0 for x  0; 500 .

29
 0  500
The expected distance and the median E X x 0.5 = 250 m,
2
2
500  0 2
the variance D( X) # 20 833.3 m ,
12
the standard deviation V(X) = D(X ) # 20 833.3 # 144.34 m,

400 300
the probability P(300 d X d 400) = F(400) - F(300) =  = 0.2.
500 500

Prob. Density Fcn. Cum. Dist. Fcn. -1 2

-1 2
Uniform Uniform

0.5 1.5

0.4

1
0.3
prob. density cum. prob.

0.2
0.5

0.1

0 0

-2 -1 0 1 2 3 -2 -1 0 1 2 3

X X

(a) (b)

Fig. 2.3 The graphs of the density function (a) and the distribution function (b) of a
uniform probability distribution.

b) The normal probability distribution N(P, V2) where P, V2 are real numbers,
V2 ! 0:
2
1 ª xP º
f x exp« », x  (- f, + f);
V 2S «¬ 2V 2 »¼

E(X) = x0.5 = x = P, D(X) = V2.

For P = 1 and V = 1, the graphs of the density function and the distribution function
are plotted in Fig. 2.4. This is the most widely used probability distribution sometimes
also called the Gauss probability distribution applied to random variables that can be

30
interpreted as the result of adding up a multitude independent influences (such as the
error of a measurement, the size deviation of a product and the like). Using the
transformation
XP
U
V

we get the standard normal probability distribution N(0;1) whose distribution function
)(u) is tabulated (see Table T1) or its values are approximated. We have

)(-u) = 1 - )(u).

For a random variable X with the normal probability distribution N(P, V2) we have

§xP·
F( x ) )¨ ¸,
© V ¹

and, for example, P(P - 3V d X d P + 3V) # 0.9973 (the so-called three-sigma rule).

E x a m p l e 2.16

What is the probability that a random variable X with the normal probability
distribution N(20, 16) will take on a value a) less than 16, b) greater than 20, c) from
12 to 28, d) less than 12 or greater than 28 ?

S o l u t i o n:
§xP·
Using the formula F( x ) )¨ ¸ and table T1 we get
© V ¹

a) P(X  16) = F(16) = )((16 – 20) / 4) = )(-1) = 1 - )(1) = 1 – 0.84135 = 0.15865 ;

b) P(X ! 20) = 1 – P(X d 20) = 1 – F(20) = 1 - )((20 – 20) / 4) = 1 - )(0) =

= 1 – 0.5 = 0.5 ;

c) P(12 d X d 28) = F(28) – F(12) = )((28 – 20) / 4) - )((12 – 20) / 4) = )(2) - )(-2)
= )(2) – (1 - )(2)) = 2)(2) – 1 = (2)(0.97725) – 1 = 0.9545 ;

d) P((X  12) › (X ! 28)) = 1 – P(12 d X d 28) = 1 – 0.9545 = 0.0455 .

31
 Prob. Density Fcn. Cum. Dist. Fcn.
11 11
Normal Normal

0.6 1

0.75
0.4
cum. prob.
prob. density 0.5

0.2
0.25

0 0

-3 -2 -1 0 1 2 3 4 5 -3 -2 -1 0 1 2 3 4 5
X X

(a) (b)

Fig. 2.4 The graphs of the density function (a) and the distribution function (b) of a
normal probability distribution

Information on further one-dimensional and multi-dimensional continuous

probability distributions can be obtained in [1], [2], [3] and [4].

32
 2. DESCRIPTIVE STATISTICS

Basic notions

When performing statistical analysis we deal with events and processes which
occur on a mass scale and can be found in a large set of individual objects such as
products or persons. We call this set a population. The objects under investigation
are called statistical items and we observe them focussing on certain properties -
statistical variables such as parameters which have outcomes or values that we
observe.

By the type of outcomes variables are either quantitative, with numerical

outcomes such as mass, length, strength, price, service life and the like, or
qualitative, which are not numeric and can only be expressed by words such as
colour, quality class, operation conditions, form, etc. If only one property is observed
we speak of a univariate variable, if more properties are ascertained at the same
time, we say that we observe a multivariate variable.

Quantitative variables are either discrete if they only take on discrete values
(number of defective products, number of faults, number of pieces, etc.) or
continuous if they assume all the values of an interval of real numbers (size of a
product, time to failure, price index and the like).

Qualitative variables are either ordinal if there is a point in ordering their

outcomes expressed in words such as quality classes or classifications or nominal
there is no point in ordering them such as colour, form, or suppliers.

Statistical methods are based on the fact that information on the population is
not taken from all its elements but rather from a subset of the population defined by
taking a sample. This is due to certain limitations such as accessibility of all the
statistical items, large size of the population, the way the information is obtained
(service life tests, wear tests, etc.), excessive costs of statistical surveys and others.
The number of statistical items in a sample is called the size of a sample. If the size

33
of a sample less than 30 to 50, we say that the sample is small, if the size is several
hundreds or thousands we say that the sample is large. This classification is of
course arbitrary and may differ depending on circumstances. A sample should be
representative, which means that it should provide information without any limitations,
and homogeneous (not affected by other factors). This can seldom be achieved with
sufficient confidence and that is the reason why we usually select the items of a
sample at random, even at the risk of the information about the whole population
contained in the sample being biased.

The following is a classification of samples according to the way they have

been selected:

 without replacement (each item may only be selected once),

 with replacement (items may be selected several times),

 intentional (typical items are selected),

 regional (the population is first divided into partitions and each partition supplies
part of the sample),

 systematic or mechanical (the population is thought to be ordered and every k-th

item is selected).

The outcomes of a variable observed or measured in the statistical items of a

sample of size n are called sample data of size n. A univariate variable X provides
univariate sample data (x1,...,xn) where xi, i = 1,..., n , is the outcome observed in the
i-th variable. Similarly, a bivariate variable (X,Y) yields bivariate sample data
((x1,y1),...,( xn, yn)) etc.

Processing univariate sample data with a quantitative variable

The primary sample data or raw scores x 1,..., x n of size n are called
ungrouped sample data. The outcomes xi may be listed in order of numerical
magnitude, which yields an array of sample data x 1 ,..., x n where x i d x i1 for

34
every i, and x 1 x min , x n x max . The interval x min ;x max is called the domain of

sample data and the number x max  x min is called the range of sample data.

If sample data are very large or if they are to be further processed (some
graphical representations or application of mathematical statistical methods) the raw
scores are grouped. Sample data are grouped by partitioning the domain into a
series of m non-overlapping intervals (usually left-open and right-closed), the so-
called classes usually of the same width h. Each class is represented by a pair x j , fj

where x j is the midpoint, x j d x j  1 and fj is the frequency (or absolute frequency) of

class j, j 1,...,m . The absolute frequency fj is defined as the number of raw scores

that lie in class j. The number fj/n is called relative frequency and sometimes it is
m
also shown as a percentage. Obviously, we have ¦f
j 1
j n.

The number of classes m is usually selected to be close to 1  3.3 log n or n.

The length of a class is then h  x max  x min / m and it should correspond to the

accuracy of the outcomes x i and for the midpoint x j to be a rounded number. For a

quantitative variable the midpoint is chosen as one of the outcomes.

j
The number Fj ¦f
k 1
k is called cumulative absolute frequency, the number

Fj / n is called cumulative relative frequency, j 1,...,m , and can be shown as a

percentage as well. Obviously, we have Fm n.

Grouped sample data is shown in a table, which is called frequency

distribution for different types of frequency:

xj x1 ... xm

fj f1 ... fm

35
 The properties of sample data are, in a concentrated form, expressed by
different measures.

The basic measures of central tendency:

The arithmetic mean (average)

1 n
x ¦ xi
ni1
for ungrouped data,

1 m
x ¦ fjx j
nj1
for grouped data.

Mathematical properties:

x y ax  b Ÿ y ax  b for arbitrary constants a, b,

x x  y xy,

x x min d x d x max ,

x x has the same unit of measurement as variable X.

Sometimes a weighted arithmetic mean

n

¦w x
i 1
i i
xw n

¦w
i 1
i

is used where w i t 0 are the weights of outcomes x i , which reflect their significance
such as accuracy.

The median

­ x § n1 · for odd n,

° ¨© 2 ¸¹
~ °
x ® x§ n ·  x§ n ·
¨ 1¸
° ¨© 2 ¸¹ ©2 ¹
°¯ for even n.
2

36
Mathematical properties:

x y ax  b Ÿ ~
y a~
x b for constants a, b,
x x min d ~
x d x max ,

x ~
x has the same unit of measurement as variable X.

The median divides the sample data into "the upper part" and "the lower part"
of outcomes xi. This is a robust measure, which, as compared to the arithmetic mean,
is little affected by extreme values. Sometimes a suitable approximation is used to
calculate the median (for example if the data is grouped).


The mode x is the number whose neighbourhood contains the most outcomes
x i , or the middle x j of the class with the largest absolute frequency fj . The mode

has the same unit of measurement as the variable X and, if it is needed, a suitable
approximation is used for calculating it.

The basic measures of variation are the following:

Variance (dispersion)

1 n 2 §1 n 2·
s2 ¦ xi  x
ni1
¨ ¦ xi ¸  x 2 for ungrouped data,
©n i 1 ¹

1 m 2 §1 m ·
s2 ¦ fj x j  x
ni1
¨ ¦ fjx j 2 ¸  x 2 for grouped data.
©n i 1 ¹

Sometimes we write s2(x). Mathematical properties:

x s 2 t 0 ,
x y ax  b Ÿ s 2 y a 2s 2 x for constants a, b,

x the unit of measurement for s 2 equals to the square of the unit of

measurement for variable X.

The more the outcomes of a variable X are scattered, the more greater is its
variance and vice versa. For calculations, sometimes an alternative formula for

37
 1 1
variance is used by replacing by . This variance calculated by this new
n n 1
n 2 2
formula equals to s ²s .
n 1

Standard deviation

s s2 .

Sometimes written as s(x). Mathematical properties:

x s t 0,
x y ax  b Ÿ s y as x for constants a, b,

x s has the same unit of measurement as variable X.

The more the outcomes of a variable X are scattered, the greater is its
standard deviation and vice versa.

Coefficient of variation

s
v .
x

Sometimes written as v(x). This is a relative measure of the variability of variable X

and it can also be shown as a percentage. It is only meaningful for a variable X with
only positive or only negative outcomes. It holds:

x v ax v x for any constant a z 0 ,

x v(x) is a dimensionless number.

The basic measure of symmetry for sample data is the coefficient of skewness

1 m 3
¦ xi  x
ni1
A for ungrouped data,
s3

1 m 3
¦ fj x j  x
n j1
A for grouped data.
s3

38
Mathematical properties:

x A ² 0 ... the majority of outcomes x i are less than x ,

x A 0 ... outcomes x i are symmetric with respect to x ,

x A ¢ 0 ... the majority of outcomes x i are greater than x ,

a
x y ax  b Ÿ A y Ax for any constant a z 0 ,
a

x A is a dimensionless number.

A number of further sample data measures exist. Sometimes the geometric

mean
xg n x 1...x n

is employed instead of the arithmetic one for some variables which describe ratios
such as volume and price indices, interest rates and the like. In special cases we use
the harmonic mean
1
§1 n 1 ·
xh ¨¨ ¦ ¸¸ .
© n i 1 xi ¹

Much valuable information on sample data can be obtained from their

graphical representation.

For univariate ungrouped or grouped sample data we can use box charts see
Fig 2.1 where the box contains the middle part of grouped data (about one half of all
the outcomes) while about a quarter of the data is placed on either side of the box.
The line on the left (on the right) corresponds to the so-called lower quartile (upper
quartile) and the perpendicular line in the middle is in the place of the median. The
height of the box is proportional to the size of the data and the line segments on both
sides represent acceptable domains for the above quarters of the data. Outcomes
beyond these line segments are considered as suspicious or extremely deviated.
There are also other modifications of this chart and other graphical tools.

39
 0 4 8 12 16
(u 1 0 0 0 )

Fig. 2.1

Two other types of charts are frequently used for univariate sorted data:
Histograms - see Fig. 2.2 - a histogram is a system of bars in Cartesian co-
ordinates where the bases if the bars are the classes and their heights
correspond to the absolute (relative, cumulative, etc.) frequencies. Frequency
polygons - see Fig. 2.3 - a frequency polygon is a broken line in Cartesian co-
ordinates connecting points whose abscissas coincide with the midpoints (or
with upper limits) of classes and their ordinates are proportional to the
frequency.

E x a m p l e 2.1

A total of 10 rollers have been measured with the following results: 5.38; 5.36; 5.35;
5.40; 5.41; 5.34; 5.29; 5.43; 5.42; 5.32. Determine the size, domain, and range,
arithmetic mean, variance, standard deviation, coefficient of variation, and median of
the sample data.

S o l u t i o n:

The data size is n = 10, the domain is <5.29; 5.43> mm and the range is
5.43  5.29 = 0.14 mm.

x = (5.38 + ...+ 5.32)/10 = 53.70/10 = 5.37 mm,

s2 = (5.382 + ...+ 5.322)/10 - 5.372 = 288.388/10 - 28.8369 = 0.0019 mm2,

s= 0.0019 # 0.0435889894 # 0.044 mm,

40
v = 0.0435889894/5.37 # 0.00811713 # 0.8117 %,
~
x = (5.36 + 5.38)/2 = 5.37 mm.

E x a m p l e 2.2

When checking the volume of beverage in a bottle for a sample of 50 bottles, the
following deviations (in ml) from the values stated on labels have been found:

1.2 2.1 1.7 0.9 0.3 2.0 -1.3 -0.1 3.2 2.8
0.8 4.4 2.9 1.2 0.0 -2.3 1.2 0.9 2.3 - 0.2
0.1 1.9 -1.9 -0.2 -1.3 1.5 0.5 2.0 -1.3 3.7
0.9 1.0 0.4 1.9 1.4 -1.3 1.6 1.4 3.1 -0.1
1.8 0.0 4.1 1.3 3.0 0.4 3.8 -0.8 3.1 0.9
Group the data, set up a frequency distribution, and design a graphical

representation. Calculate x , s2, s, x .

S o l u t i o n:

The size of the data is n = 50; xmin = - 2.3 ml and xmax = 4.4 ml, which means that
the domain is <-2.3; 4.4> ml, the range being 4.4 - (-2.3) = 6.7 ml. We choose the
number of classes to be m = 7 and the class width h = 1 (approximation of 6.7/7).
The selection of classes, their midpoints, the grouping of the data and the calculation
of absolute and relative frequencies are shown in the table below (//// stands for 5
outcomes):

j class classification fj Fj
xj

1 -2.5; -1.5 -2 // 2 2

2 -1.5; -0.5 -1 //// 5 7

3 -0.5; 0.5 0 //// //// / 11 18

4 0.5; 1.5 1 //// //// /// 13 31

5 1.5; 2.5 2 //// //// 9 40

6 2.5; 3.5 3 //// / 6 46

7 3.5; 4.5 4 //// 4 50

41
Histograms and polygons for this sample data are shown in Fig. 2.2 and 2.3. Further
calculations are, for the sake of clarity, shown in the following table:

j xj fj fj x j fj x j 2

1 -2 2 -4 8

2 -1 5 -5 5

3 0 11 0 0

4 1 13 13 13

5 2 9 18 36

6 3 6 18 54

7 4 4 16 64

¦  50 56 180

Using the table we get:

x = 56/50 = 1.12 ml, s2 = 180/50 – 1.122 = 2.3456 ml2,


s = 2.3456 #1.532 ml, x = 1 ml (the midpoint of the class with the greatest
frequency).

f 15 F 50

40
10
30

20
5
10

0 0
-3 -2 -1 0 1 2 3 4 5 -3 -2 -1 0 1 2 3 4 5
x x

Fig. 2.2

42
 f 15 F 50

40
10
30

20
5
10

0 0
-3 -2 -1 0 1 2 3 4 5 -3 -2 -1 0 1 2 3 4 5

x x
Fig. 2.3

For univariate sorted sample data with a discrete variable, usually the
following charts are used. Bar chart - see Fig. 2.4 - is similar to a histogram but there
are gaps between the bars and sometimes the bars are positioned horizontally. Pie
chart - see Fig. 2.5 - is a circle divided into sections whose perimeter corresponds to
the class frequencies. Some of the sections may be shifted in the upward or
downward direction. Different colours or types of hatching are used in these charts to
make selected pieces of information more prominent and sometimes the charts are
further geometrically and artistically modified for better presentation.

40
36
15
30
30 36
25
22 1990
1991
20 30
15 1992
1993
1994
10
22
25
0
1990 1991 1992 1993 1994

Fig. 2.4 Fig. 2.5

43
 Processing bivariate sample data with quantitative variables

The raw scores ((x1, y1),..., (xn, yn)) obtained are called ungrouped data. If we
leave out the first and the second value in each pair, we get two sets of univariate
sample data (x1,..., xn) and (y1,..., yn) respectively. Processing these sets we obtain
measures like x , y , s 2 x , s 2 y etc.

We can group bivariate sample data by grouping each of the sets of univariate
sample data x 1,..., x n and y1,..., y n where for each data set the number of classes
or the widths of classes may be different. In this way we obtain bivariate classes with
middles x j , y k and absolute frequencies f jk , j 1,...,m1 and k 1,...,m2 . The relative

frequencies fjk / n , cumulative frequencies etc. are calculated similarly.

Grouped bivariate sample data can be summarized in a crosstabulation or

contingency table for different types of frequency:

yk
xj y1 ... y m2 fxj

x1 f11 ... f1m2 fx 1

... ... ...

x m1 fm11 ... fm1 m2 fx m1

fyk fy 1 ... fy m2 n

The numbers fxj and fyk are marginal frequencies and the following formulas

hold:
m2 m1 m1 m2 m1 m2
fxj ¦ fjk ,
k 1
fyk ¦ fjk ,
j 1
¦ fxj
j 1
¦ fyk
k 1
¦¦ f
j 1 k 1
jk n.

44
 For grouped data x j , fxj , j 1,...,m1 , and y k , fyk , k 1,...,m 2 , we obtain

measures like x , y , s 2 x , s 2 y etc.

The correlation coefficient is a measure of the dependence of variables

X and Y
1 n 1 n
¦ x i  x yi  y
ni1
¦ x i yi  x y
ni1
r for ungrouped data,
sxsy sxsy

1 n 1 n
¦ f jk x j  x y k  y
ni1
¦ f jk x j y k  xy
ni1
r for grouped data.
sxsy sxsy

The numerators in all fractions define the so-called covariance cov. Sometimes we
write r(x,y) and cov(x,y).

Mathematical properties:

ac
x r(ax + b, cy + d) = r ( x, y) for constants a, b, c, d, a z 0 , c z 0 ,
ac

x  1 d r d 1 ,

x y = ax + b (a, b constants), is equivalent to r r1,

x r is a dimensionless number.

The correlation coefficient is a measure of only linear-type dependence

between X and Y. The more its value nears 1 or -1, the closer the dependence is and
the better points x i , yi can be fitted with a straight line. Its positive or negative value
corresponds to a direct or an indirect linear dependency. A value close to 0 means
either that the dependency is not linear or that X and Y are independent.

To graphically represent ungrouped bivariate sample data a scatter diagram

may be used - see Fig. 2.6 while grouped bivariate data may be shown in a
3Dhistogram - see Fig. 2.7 or in a 3D-bar chart for discrete variables X,Y.

45
Fig. 2.6

46
 Fig.2.7

E x a m p l e 2.3

A survey of costs x (CZK) and prices y (CZK) of an identical product manufactured by

ten different producers has yielded the following bivariate sample data (xi,yi):

(30.18; 50.26), (30.19; 50.23), (30.21; 50.27), (30.22; 50.25), (30.25; 50.22),
(30.26; 50.32), (30.26; 50.33), (30.28; 50.29), (30.30; 50.37), (30.33; 50.42).
Calculate x , y , s2(x), s2(y), s(x), s(y), cov, r.

S o l u t i o n:

As the data size is only n = 10 the data need not be grouped. Using the above
formulas we get:

x = (30.18 + ...+ 30.33)/10 = 30.248 CZK,

y = (50.26 + ...+ 50.42)/10 = 50.296 CZK,

s2(x) = (0.182 + ...+ 30.332)/10 - 30.2482 = 0.002096 CZK2,

s2(y) = (50.262 + ...+ 50.422)/10 - 50.2962 = 0.003684 CZK2,

s(x) = 0.002096 = 0.0457821 CZK # 0.0458 CZK,

s(y) = 0.003684 = 0.0606960 CZK # 0.0607 CZK,

47
cov = [(30.18)(50.26) + ...+(30.33)(50.42)]/10 – (30.248)(50.296) = 0.002292 CZK2,

r = 0.002292/[(0.0457821)(0.0606960)] = 0.82481996263 # 0.82482.

Judging by the value of the correlation coefficient it may be assumed that there is a
dependency between the variables that is fairly close to linear.

Processing sample data with qualitative variables

Sample data x 1,..., x n obtained are represented by a frequency distribution

where x j are all possible values of variable X expressed in words and fj are the

frequencies of these values in the original data, j 1,...,m . Measures are used only
exceptionally (variability). Bar charts and pie charts are mainly used for graphical
representation.

Sample data x 1, y1 ,..., x n , y n obtained can be grouped and summarized in a

crosstabulation or a contingency table as with quantitative variables where x j , y k

are pairs representing all combinations of the outcomes of variables (X,Y) and fjk are
the frequencies of these outcomes for j 1,...,m1 and k 1,...,m2 . Out of various
measures the most frequently used are measures of the dependence of X and Y.
3D - bar charts are used to graphically represent these data.

Exercises

E x e r c i s e 2.4

A total of ten metal parts have been machined, for each part the wasted material has
been weighed and the corresponding percentage calculated. The following data have
been obtained: 40.60; 40,29; 37.51; 38,90; 38.13; 38,15; 34.81; 37,00; 39.95; 40.43.
Calculate x , s2, s, and v.

R e s u l t: x = 38.577 %, s2 # 3.0648 %2, s # 1.751 %, v # 0.0454 = 4.54 %

48
E x e r c i s e 2.5

Calculate the domain, range, arithmetic mean, variance, and standard deviation for
the following data describing the precipitation (the amount of rain and snow fallen in
mm) in Brno from 1941 to 1960: 718.5; 492.3; 431.5; 540.5; 514.7; 584.0; 385.0;
532.0; 531.0; 578.3; 551.9; 613.6; 476.0; 661.3; 518.0; 508.5; 488.7; 494.9; 554.6;
673.5.

R e s u l t: <385.0; 718.5> mm; the range is 333.5 mm; x = 542.44 mm;

s2 # 6127.5 mm2 ; s # 78.28 mm

E x e r c i s e 2.6

In a survey conducted in 100 households selected at random the following numbers

of household members have been ascertained:

2 2 3 5 3 3 2 7 4 7 2 3 5 6 4 4 4 2 4 6 5 3 4 5 5

4 5 7 4 3 4 2 4 4 4 4 4 4 3 2 4 3 3 3 4 2 3 4 2 3

3 3 4 3 5 9 3 3 4 8 5 4 5 3 3 4 3 3 3 4 5 2 3 7 3

5 5 1 4 4 5 3 3 4 3 4 4 4 3 3 4 3 4 2 3 3 5 6 2 4

(a) set up frequency distribution tables for absolute, relative, and accumulative
frequencies

(b) calculate the average number of household members, the mode and the median.

R e s u l t: x = 3.82; x = 3; ~
x =4

E x e r c i s e 2.7

For a total of 200 parts processed by an automatic machine tool the differences from
the required size in micrometres have been measured. The following are the resulting
differences:
1.0 1.5 -2.5 0.0 -1.5 1.0 1.0 15.0 -1.0 2.0
2.0 3.0 11.0 -1.0 5.0 4.5 0.5 3.5 8.0 5.0
4.5 3.5 9.5 12.0 7.5 7.5 10.0 8.5 10.0 11.0
14.0 11.0 11.0 13.0 16.0 14.5 19.0 14.0 18.0 19.0
19.0 23.5 22.0 18.5 19.5 17.5 18.0 19.5 17.5 25.5

49
 19.5 22.0 13.5 18.5 21.5 27.5 21.0 13.5 11.5 10.0
7.5 8.5 6.5 8.5 5.5 26.0 12.5 6.5 8.5 7.5
2.5 7.0 4.5 -1.5 4.0 5.5 1.0 4.0 6.5 5.5
4.5 5.0 7.5 5.0 5.5 6.0 6.5 -3.0 5.0 3.5
-3.0 -14.0 17.0 -9.0 -3.0 -12.0 8.5 12.0 6.0 8.5
0.0 7.0 -1.0 -3.0 0.5 0.0 2.0 -4.5 2.0 -10.0
-8.5 -3.5 -11.5 -7.5 -11.5 -6.5 2.0 -11.5 -11.0 -17.5
-15.0 -15.5 1.5 -18.0 -20.0 -15.0 -3.0 -8.0 -1.0 -6.5
-8.0 -13.5 -12.0 -17.0 -10.5 14.5 10.0 9.5 7.0 0.5
21.0 10.5 5.0 0.5 4.0 0.0 0.5 3.5 9.0 2.5
2.0 7.0 7.5 3.5 7.0 4.5 -1.0 11.0 4.0 9.0
4.5 11.5 14.0 10.0 20.0 13.0 7.0 12.0 7.5 2.0
1.0 25.0 0.5 -3.0 4.5 6.0 9.5 12.5 19.0 13.0
1.5 0.5 12.0 4.0 6.5 -9.5 -8.0 -4.5 7.5 -4.0
-9.0 -9.0 2.0 -0.5 3.5 10.5 -5.5 -6.0 -6.5 -8.0

Summarize the data and use the resulting crosstabulation to calculate the arithmetic
mean, the standard deviation and the coefficient of skewness.

R e s u l t: xmin = -17.5 Pm; xmax = 27.5 Pm; h = 5.0 Pm; m = 10; x # 4.3 Pm;
s # 9.7 Pm; A # -0.102

E x e r c i s e 2.8

The below frequency distribution shows how a total of 200 workers have met the
norm. The numbers in the upper line are the midpoints of percentage classes:

xj 85 95 105 115 125 135 145 155 165 175

Fj 4 21 65 39 24 17 12 9 7 2

Use the frequency distribution to calculate the arithmetic mean, mode, median,
standard deviation, and coefficient of skewness.

R e s u l t: x = 117.9 %; x = 105 %; ~
x = 115 %; s2 # 380 %2; A # 0.92

50
E x e r c i s e 2.9

In a statistical survey made by an insurance company each person has been asked
about the bonus they are paying. The following is the resulting frequency distribution.
The upper line shows the bonuses in CZK:

xj 390 410 430 450 470 490 510 530 550 570

Fj 7 10 14 22 25 12 3 3 2 2

Calculate the arithmetic mean, mode, median, variance, standard deviation,

coefficient of skewness, and coefficient of variation.

R e s u l t: x = 457.4 CZK; x = 470 CZK; ~
x = 450 CZK; s2 = 1493.24 CZK2;
s # 38.64 CZK; A # 0.52; v # 8 %

E x e r c i s e 2.10

Calculate the measures for the following bivariate sample data:

xi 18 19 20 21 22 22 25 26 26 26 27 28 29 30 31 33

yi 26 23 29 27 31 25 22 32 32 33 38 29 36 37 41 42

R e s u l t: x # 25.19; y # 31.44; s2(x) # 18.777; s2(y) # 35.246;

s(x) # 4.33; s(y) # 5.94; cov # 20.980; r # 0.8155

E x e r c i s e 2.11

Calculate the measures for the following bivariate sample data:

xi 2 4 4 5 6 8 10 10 10 10

yi 1 2 3 4 4 4 5 5 5 6

R e s u l t: x = 6.9; y = 3.9; s2(x) = 8.49; s2(y) = 2.09;

s(x) # 2.91; s(y) # 1.45; cov = 3.89; r # 0.9235

51
E x e r c i s e 2.12

The following contingency table summarizes last year's (x) and this year's (y) prices
of shares in thirty companies selected at random. Find the average prices of shares
and the correlation coefficient.

yk
1001 - 2000 2001 - 3000 3001 - 4000
xj

501 - 1000 3 6 0

1001 - 1500 5 8 2

1501 - 2000 0 1 3

2001 - 4000 0 1 1

R e s u l t: x = 1283.8 CZK; y = 2433.8 CZK; r = 0.4232

Questions

1. Describe the types of variables and show examples.

2. What is a sample, what are its properties and how it is done?
3. Define sample data and show how it is related to the parent population.
4. Describe the way univariate sample data with a quantitative variable are grouped.
5. What are the measures of central tendency for univariate sample data with a
quantitative variable, what are their properties and what is their significance?
6. Show the measures of variation for univariate sample data with a quantitative
variable, their properties and their significance.
7. Describe types of graphical representation of univariate sample data with a
quantitative variable.
8. Show the measures for bivariate sample data with quantitative variables.
9. Describe the way bivariate sample data with quantitative variables are
summarized and the types of their graphical presentation.
10. Describe the way sample data with qualitative variables are processed and
graphically presented.

52
 3. ANALYSIS OF TIME SERIES

Fundamentals

Two main categories of statistical information exist: cross sections and time
series. The economists often estimate the consumption by relating the consumers'
costs to the national product or analyse in detail the distribution of consumption at
one particular point of time (cross sections). This approach has a broader
significance for the practice but is not sufficient if we are interested the dynamic of an
event and in particular changes over time.

The basic tool employed to study of the dynamic an event is an analysis of its
past development, which helps us grasp the existing laws and to estimate its future
development.

A time series is obtained if the data on a particular event over time are
arranged in order of increasing time. A well-established time series that can be used
for an analysis must meet the following requirements.

- the data must be arranged in order of increasing time,

- the data items must be comparable:

a) the same period of time over which the data has been acquired,

b) the same data definition (units of measurement, uniform data

collection method).

Failure to comply with any of the above conditions may result in erroneous
conclusions.

From the statistical point of view a time series is a sequence (y1,...,yn) of the
observed values of a statistical variable Y where the index i corresponds to the time ti
or to the i-th interval ending at ti, ti < ti+1, i = 1,...,n. Sometimes we write yt instead
of yi. Graphically, the time series is mostly represented by a graph in the Cartesian

53
system of co-ordinates with the indices i or times ti as abscissas and the values yi as
ordinates. Fig 3.1 shows an example of a time series.

MONTHLY CHAMPAGNE SALES

15

12

9
Sale
6

0
0 12 24 36 48 60 72 84
Time

Fig. 3.1

If time series are related to periods of time, they are called interval time series,
if they refer to points of time, they are called point time series.

Interval time series are composed of indexes measured for fixed time intervals
such as an hour, day, month, or year, etc. They are characterised by the following
features:
- the data items express quantities,
- they are dependent on the length of the time interval,
- the sum of the data is meaningful.

Point time series contain data that are related to a fixed point in time. They are
characterised by the following features:

- the data express a level or a condition of the variable under investigation,

- there is no point in adding up all the data items.

54
 To analyse a time series correctly certain differences must be taken into
consideration that follow from the different character of time series data and from
their significance.

The following is a classification of time series:

1) absolute quantities: a) interval

b) point

2) derived quantities: a) sums: D) cumulative

E) moving sums

b) averages: D) cumulative averages

E) moving averages

c) quotients

Interval and point time series

Interval series

It is typical of interval time series that they are related a fixed time interval and
as such are affected by the length of the interval. The following are the most common
interval quantities: production volume, retail sales, revenues, wages and salaries,
man-hours, number of children born within a certain period, etc.

Point time series

The quantities used for setting up a time series typically do not refer to an
interval but rather to a time point. This may be the first or the last day of a period, an
arbitrary but fixed day or moment. The number of inhabitants, workers, the amount of
fixed assets, etc may exemplify data of this type. Such data show an instantaneous
condition of the event in question. We use the following numerical characteristic
called the chronological mean to aggregate the data.

55
 Given the values of a point index (outcomes of the observed variable Y)

y1, y2,..., yn

for n time points

t1, t2,..., tn,

we can calculate the averages for the following points

t1 and t2, t2 and t3,..., tn-1 and tn .

Using these partial averages we now calculate the average for all the aggregate point
values

y1  y 2 y2  y3 y n1  y n
y1 , y2 ,..., yn1 .
2 2 2

The number dividing the sum of the above partial averages will be one less than n.

If the intervals between the individual values of a point time series are of equal
length, the chronological mean is calculated as follows. Denoting the distances
between the members of a point time series by

d1, d2,... , dn-1,

we have

d1 = d2 =...= dn-1,

and the chronological mean is

y1  y 2 y 2  y 3 y  yn
  "  n1
y chr 2 2 2 .
n 1

After some manipulations we get

0.5y1  y 2  y 3  "  yn1  0.5yn

y chr .
n 1

If the distances between the neighbouring members of a point time series are
not equal, the calculation is similar. We reduce the different lengths of time
d1, d2,..., dn to one value. We use their weighted average to do this:

56
 y1 y2 y3 yn-1 yn

t1 t2 t3 tn-1 tn
d1 d2 dn-1
The chronological mean is then

y1  y 2 y  y3 y  yn
d1  2 d2  "  n1 dn1
y chr 2 2 2
d1  d2  "  dn1

which yields

y1d1  y 2 (d1  d2 )  ...  y n1(dn2  dn1 )  y ndn1

y chr .
2 d1  d2  ...  dn1

E x a m p l e 3.1

The following are data on the numbers of employees of a company during the
calendar year:

1st Jan 3 500 employees

1st Apr 3 425 employees

1st Jul 3 430 employees

1st Oct 3 390 employees

1st Jan 3 350 employees

Calculate the chronological mean of the time series expressing the numbers of
employees.
Solution:

We use the above formulas to calculate the chronological mean, where y1 = 3 500,
y2 = 3 425, y3 = 3 430, y4 = 3 390, y5 = 3 350 and d1 = d2 = d3 = d4. Substituting
these values we get

(0.5)(3 500)  3 425  3 430  3390  (0.5)(3 350)

y chr 3 417.5 .
4

The chronological mean of the numbers of employees of the company is 3 417.5 for
the given year. For practical use we can round off the value to 3 418.

57
E x a m p l e 3.2

A company keeps an inventory of stock. The total figures are in Czech Korunas. The
data for the following dates are available:

1st Jan 20.523 million

1st May 16.100 million

1st Oct 17.230 million

1st Jan 21.432 million

Calculate the average yearly stock in the company.

Solution:

We use the formula for the calculation of the chronological mean, where y1 = 20.523,
y2 = 16.100, y3 = 17.230, y4 = 21.432 and

1st Jan 1st May 1st Oct 1st Jan

d1 = 4 d2 = 5 d3 = 3

Substituting the data we get

(20.523)( 4)  16.100 4  5  17.230 5  3  (21.432)(3)

y chr # 17.880 .
2 453

The average yearly stock in the company was 17.880 million CZK.

Special types of time series

Cumulative time series

Cumulative time series behave like increasing sums. A cumulative time series
is formed by gradually adding up the values of a given variable starting from a fixed
point. This method is employed for example when monitoring indexes for a certain
period of time such as a month or a year. Cumulative values are of considerable help

58
in matters of strategic decision making. In the following example use of this method is
demonstrated.

E x a m p l e 3.3

Using the following data on the production volume for each month of the year
analyse the real production for each month and for the whole year.

Production (thousands of tons)

Monthly values Cumulative values

Month
Plan Production (%) Plan Production (%)

January 36.2 36.5 100.8 36.2 36.5 100.8

February 36.5 35.8 98.1 72.7 72.3 99.4

March 35.1 35.8 102.0 107.8 108.1 100.3

April 34.2 35.1 102.6 142.0 143.2 100.8

May 33.0 33.2 100.6 175.0 176.4 100.8

June 33.0 32.1 97.3 208.0 208.5 100.2

July 33.0 31.2 94.5 241.0 239,7 99.5

August 32.5 31.0 95.4 273.5 270.7 99.0

September 32.7 32.3 98.8 306.2 303.0 99.0

October 34.6 33.4 96.5 340.8 336.4 98.7

November 36.8 36.6 99.5 377.6 373.0 98.8

December 38.0 38.1 100.3 415.6 411.1 98.9

From the tabulated values it can be concluded (cf. December) that, as compared with
the planned value 415.6 thousand tons, the yearly total was 411.1 thousand tons,
which is 98.9 % of the plan.

59
 Time series of cumulative averages

Time series of cumulative averages are derived from interval series. These
series show how the cumulative averages approach the total average over the given
period of time, which is equal to the last value.
This method is used for example to record the costs in quality control. It is
based on a cumulative time series where the values are divided by the number of
periods over which it has been accumulated. We will use the data from Example 3.3
to demonstrate this. The quantities are shown in thousands of tons.

E x a m p l e 3.4

Cumulative production and cumulative averages (thousands of tons):

January 36.5 36.5 / 1 = 36.50

February 72.3 72.3 / 2 = 36.15
March 108.1 108.1 / 3 # 36.03
April 143.2 143.2 / 4 = 35.80
May 176.4 176.4 / 5 = 35.28
June 208.5 208.5 / 6 = 34.75
July 239.7 239.7 / 7 # 34.24
August 270.7 270.7 / 8 # 33.84
September 303.0 303.0 / 9 # 33.67
October 336.4 336.4 /10 = 33.64
November 373.0 373.0 /11 # 33.91
December 411.1 411.1 /12 # 34.26

Cumulative series of moving sums

Series of moving sums are interval-type time series. In practice we usually

calculate moving yearly totals using data for individual months. As a rule, we use
data collected over a minimum of two years. These series are created by gradually
adding up consecutive values. A series of moving sums is suitable for comparing
development trends over two different periods.

60
E x a m p l e 3.5

1994 January 36.5 36.5+35.8+33.2+31.2+32.3+36.6 = 205.6

February 35.8 35.8+33.2+31.2+32.3+36.6+36.8 = 205.9
May 33.2 .............. = 206.2
July 31.2 .............. = 207.0
September 32.3 .............. = 207.8
November 36.6 .............. = 209.0

1995 January 36.8 .............. = 209.6

March 36.1
May 34.0
July 32.0
September 33.5
November 37.2

The values for 1994 and 1995 clearly show that the production trend is increasing.

Time series of moving averages

Moving averages are related to the calculation of moving series. Moving

averages are calculated by dividing moving sums by the number of period over which
the sum has been made. A series of moving averages smoothes possible seasonal
influences of current values. More information on moving averages can be found in
[1], [3], [4], and [5].

Development of time series

The simplest numerical characteristics used to analyse time series are the
absolute and relative measure of growth and decline. An analysis of absolute and
relative measures of growth enables decisions necessary for the selection of a
function used for smoothing a time series. For the following methods, we will always
assume that the neighbouring boundaries or midpoints of the time intervals are
equidistant.

61
 Absolute measures of growth provide absolute comparison between the
members of a time series. The following measures are used:

absolute increment (difference)

Gi y i  y i  1, for i = 2, 3,…, n,

mean absolute increment

1 n y n  y1
G ¦ Gi
n1i 2 n 1
,

which is calculated as the simple arithmetic mean of all absolute increments.

If all average increments (also called first differences G(i1) ) are close to a
constant, the time series has a linear trend, which can be expressed in the form of a
straight line. Second differences G(i 2 ) are obtained by subtracting two neighbouring

first differences. Sometimes also their arithmetic mean G ( 2 ) is calculated. If the

second differences of a time series are all close to a constant, the time series may be
represented by a parabola. Third differences are calculated as the differences of two
neighbouring second differences. Further differences are established in a similar
way. If the differences of order k are almost constant, the corresponding time series
may be represented by a k-th order polynomial.

The rate of development (growth or decline) of the values of a time series is

characterised by relative increments calculated as the ratio of the first difference of
the i-th period to the current value of the (i -1)th period. The development rate may be
expressed by the following characteristics:

coefficient of growth

yi
ki for i = 2, 3,..., n,
y i1

62
 coefficient of increment

Gi
Ni k i  1 for i = 2, 3,..., n,
y i1

average coefficient of growth

n
k n 1 k 2 k 3 ...k n n1
–k
i 2
i ,

which is calculated as the geometric mean of coefficients of growth.

The average coefficient of growth may also be calculated as the (n - 1)-th root of the
quotient of the first and the last current value in the given time series

y 2 y 3 y 4 y n1 y n yn
k n1 " n1 .
y1 y 2 y 3 y n2 y n1 y1

Coefficients of growth and increment as well as the corresponding average

values are sometimes expressed as percentages: ki100%, Ni100%, k100% . If the
growth coefficients ki of a time series vary slightly around a constant, the trend of the
series is approximately exponential.

E x a m p l e 3.6

The GDP figures (in thousands of millions of CZK) in the Czech republic between
1990 and 1996, recalculated for fixed prices, is given in the table below.

Year 1990 1991 1992 1993 1994 1995 1996

GDP 564 721 859 1015 1192 1338 1579

Determine the average GDP, absolute yearly increments, the average yearly
increment, second differences, the average second difference, the coefficient of
growth, and the average coefficient of GDP growth.

63
Solution:

Part of the results can be found in the table where t as the time variable is used
instead of i:

t yt Gt Gt(2) kt kt100% Nt Nt100%

1990 564 --- --- --- --- --- ---

1991 721 157 --- 1.2784 127.84 0.2784 27.84

1992 859 138 -19 1.1914 119.14 0.1914 19.14

1993 1015 156 18 1.1816 118.16 0.1816 18.16

1994 1192 177 21 1.1744 117.44 0.1744 17.44

1995 1338 146 -31 1.1225 112.25 0.1225 12.25

1996 1579 241 95 1.1801 118.01 0.1801 18.01

6 7268 1015 84 --- --- --- ---

The average yearly GDP is

7268
y # 1038.2857 # 1038.3 thousands of million CZK.
7

Absolute increments Gt, second differences Gt(2), coefficients of growth kt and

coefficients of increment Nt are shown in the table. Hence we see that 241 thousands
of million CZK was the greatest value of the GDP reached in 1996, and in 1992 it
dropped to a minimum of 138 thousands of million CZK. However the largest relative
growth of the GDP was attained in 1991 (the coefficient of growth being 1.2784 or
127.84 %) and the least relative growth of the GDP was recorded in 1995 (the
coefficient of growth being 1.1225 or 112.25 %). Further the table shows that the
greatest absolute acceleration of the GDP (the greatest positive second difference)
was reached in 1996 and the greatest slow-down of the GDP (the least negative
value of the second difference) rate was recorded in 1995. The average yearly
absolute increment of the GDP is

1075
G = 169.1666…#169 thousands of million CZK.
7 1

64
The average yearly growth coefficient for the GDP is

1579 6
k 71 # 2.7996454 # 1.1872 or 118.72 %.
564

Hence the average yearly growth coefficient for the GDP is 0.1872 or 18.72 %. The
calculation of the average coefficient of growth that uses the arithmetic mean may
be misleading but unfortunately this often occurs in economic applications. The
average yearly second difference of the GDP (in thousands of millions of CZK) is

84
G ( 2) = 16.8 ! 0,
72
which means that the overall growth of the GDP is increasing.

Time series analysis

Time series is the measurement of a variable over time. The following major
components, or movements, of time series may be identified:
- trend (long-term influence),
- periodic influences (recurring regularly) affecting the values of a time series,
- irregular influences (occurring at random, forecasting is difficult).

The trend of a time series

The trend is an important component of time series. It may be linear or any

other non-linear function of time. The trend is the general movement over a long
period of time. When using time series analysis in economy, we are interested in the
trend both in terms of the present situation and a forecast of its future development. A
number of methods have been devised and computer programs have been written to
express the trend of time series [1], [3], and [4], [5] a [6]. The basic methodology is
described below in the paragraph on time series smoothing.

Periodic influences

Periodic influences account for periodic variations of time series over time. The
length of the periods varies as well and can be used for further subdivision of periodic

65
influences as follows:

- cycles (wavelike repetitive movements fluctuating about the trend of the series),

- seasonal variation (repetitive fluctuating movements occurring within a time

period of one year or less),

- short-term influences (fluctuations with regular periods such as a day in the

week, a week in the month and the like).

A number of methods have been devised and computer programs have been written
to express the periodic changes of time series [1], [3], [4], [5] and [6].

Irregular influences

The irregular component of a time series refers to random movements around

its trend or seasonal component. We consider these movements as interfering
components. They can be represented by random variables and may be described
by statistical methods [1] and [2]. Some of the methods are treated in Chapter 5.

Time series decomposition

A time series may be thought of as the result of its trend component Tt, periodic
component Pt and random component Et. The periodic component may further be
decomposed into a cyclic component Ct and a seasonal component St. The
decomposition of a time series is mostly based on an additive model

yt = Tt + Pt + Et , or yt = Tt + Ct + St + Et

or a multiplicative model

yt = Tt Pt Et , or yt = Tt Ct St Et .

Time series smoothing

When analysing the trend component of a time series, we try to identify the
influence of those factors that are stable and determine the trend. Graphically, this
corresponds to drawing a curve that best fits the time series trend when plotted. Such

66
a curve may be obtained by graphically, mechanically, or analytically smoothing the
time series.

For graphical smoothing the time series is plotted in a graph (Fig. 3.1) and its
trend is estimated (smoothed) graphically. This method can only serve as a guideline
and sometimes it may be misleading.

Mechanical smoothing of a time series is based on moving sums. When

moving sums are divided by the number of periods, moving averages are obtained
with values mostly close to the original values but devoid of their seasonal variations.
Thus the moving average curve tends to be smoother than the original one.
Moreover, it becomes more monotonous (and shorter) as more periods are taken as
the basis to calculate the moving sums. A major advantage of this method is its
simplicity and the fact that it informs us well on the development trend of the time
series without interfering seasonal and cyclic fluctuations.

Analytical smoothing of a time series is based on the assumption that a

function of time exists that approximates the values of the time series. We try to
select a function that best suits its development. The selection can be guided by a
graphical representation of the time series empirical values or by calculating its first
(second or other) differences. The following is the general form of a smoothing
function:

yct f ( t)  e t ,

where: yct - denotes the smoothed values of the dependent variable,

t - denotes the independent time variable,

et - denotes the so-called residual component.

The function f(t) ought to render the trend of the time series correctly, that is,
to smooth the time series as well as possible. Linear, parabolic, and exponential
functions are the ones most frequently used. Generally, any function may be used -
more details can be found in Chapter 5.

The most frequently used method is linear smoothing if the trend of the time

67
series appears to be linear. The function f(t) has the form

y ct b 0  b 1t ,

and the parameters bo and b1 are determined from the so-called system of normal
equations
T T
b0 T  b1 ¦ t ¦y t ,
t 1 t 1

T T T
b0 ¦ t  b1¦ t ¦ yt t . 2

t 1 t 1 t 1

The first coefficient b0 determines the point at which the smoothing straight
line intersects the y-axis. It is interpreted as the smoothed value of the time series in
period zero. The second coefficient b1 is the slope of the straight line and expresses
the actual trend. It determines the change of the smoothed values y't for a unitary
change of t or the average change of the original values yt when t is increase by
one. We can test the suitability of the smoothing function by looking at the plotted
diagram or by calculating the correlation coefficient of the pairs (t, yt) or the sum of
the squared differences ¦(yt - y't)2.

This method may be simplified if we shift the time variable for the sum of its
shifted values to be equal to zero. This can be achieved by shifting the origin (0) to
the central period, that is, by decreasing t by its mean value
T1
t .
2
Thus, instead of t, we consider the variable t´ = t - t . By this transformation the
terms in the system of normal equations with 6 t turn to zero, which yields the
following explicit formulas
T
T

¦y t ¦ y t tc
bc0 t 1
, b1c t 1 .
T T

¦ tc2
t 1

For the original coefficients we have

b0 = b´0 - b´1 t , b1 = b´1 .

68
E x a m p l e 3.7

Determine the trend component of the time series representing the development of
the gross domestic product in the Czech Republic from 1990 to 1996 as shown in
Example 3.6.

Solution:
As the original time variable t takes on the values 1990, 1991,…, 1996, we will use
the transformation t´ = t - 1993 since then we have

t = (1990 + 1991 + … + 1996)/7.

The number of periods T = 7 so that we get

564  721  "  1579 7268

bc0 1038.2857 # 1038.3 ,
7 7

( 3)(564)  ( 2)(721)  "  (3)(1579) 4612

b1c 164.71428 # 164.7 .
( 3) 2  ( 2) 2  "  3 2 28

The original coefficients being

b0 = 1038.2857 – (164.71428)(1993) = -327 237.28 # -327 237.3 ,

b1 = 164.71428 # 164.7 .

Hence we get the following straight line, which smoothes the time series

yct 327 237.3  164.7 t .

c
For t = 1993, say, we obtain y1993 327 237.3  (164.7)(1993) 1009.8 , which is in

good correspondence with the real GDP y1993 = 1015. Also the value b1 = = 164.7 is
close to the average increment G = 169 from Example 3.6.

Exercises

E x e r c i s e 3.8

The below data describe the amount of fixed assets in a company over a calendar
year (the accounting value):

69
 1st Jan 101.230 million CZK 1st Aug 100.250 million CZK
1st Mar 105.100 million CZK 1st Dec 99.800 million CZK
1st Apr 105.500 million CZK 1st Jan 103.150 million CZK

Calculate the average fixed assets for the calendar year.

R e s u l t: y chr = 102.05875 million CZK

E x e r c i s e 3.9

The figures displayed in the table show electricity consumption in Czechoslovak

industry between 1967 and 1972 in thousands of millions of kilowatt-hours.

Year 1967 1968 1969 1970 1971 1972

Consumption 27.6 29.1 30.3 31.8 33.6 35.4

Determine the average yearly electricity consumption, absolute yearly increments,

the average yearly increment, second differences, the average second difference,
the coefficient of growth, the coefficient of increment, and the average coefficient of
growth of electricity consumption.
R e s u l t: y = 31.3 ; G 1.56 ; G ( 2 ) 0.075; k # 1.051

E x e r c i s e 3.10

Smooth the average fixed assets figures (in millions of CZK) shown in the table below
for the years 1978 to 1985 using the linear smoothing method and calculate a
forecast of the average fixed assets for 1987 assuming that the trend of the time
series remains does not change.

Year 1978 1979 1980 1981 1982 1983 1984 1985

Amount 675.0 681.4 684.0 689.6 690.8 698.2 706.0 712.0

R e s u l t: yct c
9352.193  5.069t , y1987 719.91 mill. CZK

E x e r c i s e 3.11

A transport company recorded the following average numbers vehicles in its fleet:

70
 1993…54 , 1994…63 , 1995…69 , 1996… 72.

The company's real figures as of particular dates of 1997 are shown in the below
table:

Date 1st 20th 15th 30th 25th 31st

Jan Mar Apr Jul Sep Dec

Number 70 66 71 80 82 90

Determine the average number of the company's vehicles for 1997, the average
yearly coefficient of growth and characterise the trend of the development from 1993
to 1997 using a linear function.

R e s u l t: y chr # 77 (weighted yearly average); k # 1.093 (average yearly increase

by 9.3 %); yct 10 905.5  5.5t

Questions

1. Define a time series and show examples.

2. What is the classification of time series. Show examples of each type.
3. How is the average of a time series calculated?
4. What characteristics describe movements of a time series?
5. Describe the components of a time series and its decomposition.
6. What methods are used to smooth a time series?

71
 4. INDEX NUMBERS

Basic notions

Index numbers are relative statistical measures that express change in

magnitude of a quantitative variable or a group of variables describing the behaviour
of one or several items over a period of time or as influenced by a factor. They
usually take the form of a fraction where the numerator is the value of the variable for
the current period and the denominator is its value for the base period. Depending on
the character, the construction complexity, and the influence of the quantities
observed, indexes are either simple or composite.

Simple indexes measure the relative change from the base period for a single
item or for a group of homogeneous items. In the former case they are called single
indexes (describing such quantities as the price or the amount of a single product)
and in the latter case they are called group indexes (when used to measure the
change in one variable (such as the price or quantity) for a group of homogeneous
items. As opposed to that, composite indexes measure the relative change from the
base period in a group of inhomogeneous items or an aggregate (such as a bundle of
commodities of different types).

Indexes that measure the number of items, the production volume and the like
are called quantity indexes and are denoted by q while those that measure such
quantities as price or intensity are called value indexes and are denoted by p. Among
the indexes of the first group the most frequently used are volume indexes while
those in the second group are usually price indexes.

Simple index numbers

Single indexes

The single index iq for a quantity or ip for a value is given by

72
 q1 p1
iq or ip ,
q0 p0

respectively where the numerator corresponds to the current period and the
denominator corresponds to the base period. The correct selection of the base period
is important. Those values that best represent the outcomes of the variable should be
chosen as the base. Sometimes it is best to use the average of several values. When
calculating and interpreting index numbers, we must ensure comparability of the
periods and a factual agreement of the aspects concerned otherwise the expressive
power of the index could be significantly impaired.

E x a m p l e 4.1

The production volume of a steel works reached 2780 tons in 1994 when the price of
steel was 8750 CZK and in 1995 the production volume rose to 2950 tons with a
price of 9690 CZK. Calculate indexes for the production and price of steel.

2950 9690
Solution: iq # 1.061 106.1% , ip # 1.107 110.7%
2780 8750

Group indexes

While single indexes are used to analyse single items such as the quantity of
cement produced by one plant, group indexes are related to a group of similar items
such as the cement production volumes for a group of plants. The distinction
between single and group indexes is of great importance. For group indexes, the
comparability of periods, facts and composition plays an important role. For a
quantity variable, the group index is given by

¦q
i
(i)
1
iq .
¦q
i
(i)
0

E x a m p l e 4.2

The production figures for four cement production plants are given in the following
table:

73
 Plant January (0) February (1) March (2)

A 2300 t 2450 t 2390 t

B 5210 t 4800 t 5100 t

C 8100 t 8600 t 9000 t

D 6250 t 6250 t 5900 t

Total production 21860 t 22400 t 22090 t

If we consider all the plants as one unit, the group indexes for a quantity variable are
as follows (in fact they are what will be later called chain production indexes):
22 400 22 090
iq1 # 1.025 , iq2 # 0.986 .
21 860 22 400

Group indexes may be further specified as variable composition indexes, fixed

composition indexes, and structure indexes. A group index describing changes in
values (such as average prices) is called a variable composition index and is given
by

¦p q
i
(i) (i )
1 1

p1 ¦q i
(i )
1
i var .comp. .
p0 ¦p q
i
(i)
0
(i )
0

¦q i
(i )
0

E x a m p l e 4.3

The following table gives a summary of prices in a company:

No. of items Price per item

Supply value in period
in period in period
Supply (in thousands of CZK)
Base Curr. Base Curr.

q0 q1 p0 p1 p0q0 p1q1 p1q0 p0q1

Contractual 8000 8000 12.- 13.2 96 105.6 105.6 96

Surplus 500 4600 30.- 30.- 15 138 15 138

Total 8500 12600 ----- ----- 111 243.6 120.6 234

74
The value under investigation is the price. Price changes in individual types of supply
may be measured by single indexes for contractual (1) and surplus (2) supplies:

p1(1) 13.2 p1( 2) 30

1.100 , 1.000 .
p(01) 12 p(02) 30

For contractual supplies the price has risen by 10% while for surplus supplies it has
remained the same. The average price for the base period is

p0
¦p q (i )
0
(i)
0 111 000
# 13.06 CZK
¦q (i )
0 8 500

and the average price for the current period is

p1
¦p q (i) (i)
1 1 243 600
# 19.33 CZK,
¦q (i)
1 12 600

which yields a variable composition index of

19.33
i var .comp. # 1.480 148% .
13.06

Composite indexes

The basic property of composite (aggregate) indexes is that they can be used
to measure changes in quantities and values of inhomogeneous variables. If, for
example, the prices of consumer goods have changed, we would like to know the
percentage of the drop in prices of products as a whole. Since we have to deal with a
range of different types of goods, the average price of a unified item cannot be used.

The change may be determined by considering it to be the sum of values

multiplied by the corresponding quantities. Thus in the above example the total value
will be the result of two variables:

a) value variable (such as price, total costs, and labour intensity),

b) quantity variable (a carrier of the above values).

75
Then the so-called composite value index (retail trade turnover) is given by

¦q i
(i ) ( i)
1 1 p
Ih .
¦q i
(i ) ( i)
0 0 p

To determine the influence of only one of the variables, the influence of the
other must be eliminated, which is done by fixing it at a given constant level in each
of the aggregates that are being compared. A constant level for a value variable (p)
or a quantity variable (q) to calculate the index may be achieved in two ways: by
fixing it at the level of either the base period or the current period. Accordingly, we get
the following indexes:

¦q (i)
1 p(0i)
the Laspeyres composite index for quantity ILq i
,
¦q
i
(i)
0 p(0i)

L
¦q
i
( i ) (i )
0 1 p
the Laspeyres composite index for value Ip ,
¦q
i
(i)
0 p(0i)

P
¦q i
(i) (i)
1 1 p
the Paasche composite index for quantity I q ,
¦q i
(i) (i )
0 1 p

P
¦q i
(i) (i)
1 1 p
and the Paasche composite index for value I p .
¦q i
(i)
1 p(0i)

The following relationships are easily established between the composite

indexes for quantity and value and the composite value index

Ih ILpIPq ILqIPp ,

since we have

¦ q1p1 ¦ q0p1 ¦ q1p1 ¦ q1p 0 ¦ q1p1

.
¦ q0 p 0 ¦ q0 p 0 ¦ q0 p1 ¦ q0p 0 ¦ q1p 0

76
 Neither the Laspeyres nor the Paasche index expresses a change in a
satisfactory manner since a change in p or q between the base period and the
current one may cause a change in q or p respectively. For example a change in
prices may influence the consumption and vice versa. To eliminate this drawback,
we use sometimes the Fisher ideal index

IFq ILq IPq or IFp ILp IPp .

However, not even the Fisher index does reflect the changes sufficiently (it is only a
compromise between the Laspeyres and the Paasche index) and therefore further
indexes are used as well [1].

E x a m p l e 4.4

The following table contains sales figures for products A, B, C in a trading company:

Items sold Item retail price

Product Base Current Base Current

period period period period

q0 q1 p0 p1

A 1 000 1 200 60 69

B 6 000 4 500 10 11

C 8 000 9 000 8 7

Determine: a) the growth of the total retail trade turnover in the company for the
given period,
b) the growth of the sales volume,
c) the change in the price of the commodities sold.

Solution:

The following table contains auxiliary calculations:

77
 Product iq ip q0p0 q1p1 q0p1 q1p0

A 1.200 1.150 60 000 82 800 69 000 72 000

B 0.750 1.100 60 000 49 500 66 000 45 000

C 1.125 0.875 64 000 63 000 56 000 72 000

Total --- --- 184 000 195 300 191 000 189 000

Hence the retail trade turnover index

195 300
Ih # 1.0614 = 106.14%.
184 000

The above index tells us that the retail trade turnover in the given company rose by
6.14 %. The Laspeyres index for the sales volume is

189 000
ILq # 1.0272 = 102.72%
184 000

and the Paasche index for the prices is

195 300
IPp # 1.0333 = 103.33%.
189 000

The sales volume rose by 2.72% and the price of the commodities sold rose by
3.33%. The following relationship may be established for the two indexes

Ih ILq IPp # (1.0272)(1.0333) # 1.0614 .

Further indexes are calculated in a similar way:

195 300
IPq # 1.0225 = 102.25%
191 000

and
191 000
ILp # 1.0380 = 103.80% .
184 000

Index numbers and absolute quantities

Indexes express only relative changes in observed variables. They are not
sufficient to analyse the development to the full extent. For this reason we must know

78
also the absolute value of the change expressed by an index. To do this we proceed
in different ways for quantity and value variables. For quantities, apart from the index

q1
iq ,
q0

we are also interested in the absolute value of the quantity under investigation, which
is given by

q1  q0 .

In value variables, the difference between the numerator and denominator of the
index fraction only indicates the increase (decrease) of the level.

Basic and chain indexes

When analysing real events we must sometimes set up a whole series

(sequence) of indexes for several subsequent periods. The following two indexes are
mainly used.

a) One period is taken for a base with the value y0 of the variable to be analysed and
the ratios of other outcomes yn in current (other) periods to this base period are
calculated. In this way we obtain what is called basic indexes or constant base
indexes,

yn
in / 0 , n = 0, 1, 2, …
y0

b) A given period is always compared with the previous one, yn being divided by yn-1.
In this way we obtain chain indexes or moving base indexes,

yn
in / n1 , n = 1, 2, …
y n1

Chain indexes must pass the so-called intercalation test

i m / k in / m in / k .

We use both basic and chain indexes to set up time series to be processed by
methods described in Chapter 3. For example, the geometric mean of chain indexes
is the average index that expresses the same relative change in the given variable

79
between individual periods of equal length. Note that, in this case, we cannot use the
arithmetic mean.

E x a m p l e 4.5

Calculate the basic and chain indexes for the data in the following table.

Period Production Denotation

January 75 000 t q0

February 75 250 t q1

March 81 000 t q2

April 82 100 t q3

Solution:

Basic indexes: i0/0 = 1.0000; i1/0 # 1.0033; i2/0 = 1.0800; i3/0 # 1.0947

Chain indexes: i1/0 # 1.0033; i2/1 # 1.0764; i3/2 # 1.0136

Exercises

Exercise 4.6

The beer production of a brewery for individual types of beer with the average yearly
prices in 1996 and 1997 is given in the table where q is the quantity (hl) and p is the
price (CZK/hl):

Type 100 110 120

Period q p q p q p

1996 8350 1000.- 2560 1200.- 3870 1450.-

1997 9460 1100.- 3440 1350.- 2800 1600.-

Since the products are homogeneous, calculate both single and group indexes

80
related to the base year 1996 and interpret them in terms of absolute changes in
price, quantity, and value of the beer production.

Exercise 4.7

Using single indexes, composite indexes (of the Laspeyres, Paasche, and Fisher
type) for both value and quantity, and the cost index, calculate changes for a "small"
market basket of a typical four-member family. The average retail prices p (CZK/kg or
CZK/l) and the quantities of purchased food q (kg or l) are shown in the following
table. Interpret the resulting indexes in terms of absolute changes in the price of the
basket.

meat bread pastry soft drinks

Food

Period q p q p q p q p

May 94 12 90.- 20 11.- 8 21.- 30 5.-

May 95 11 110.- 18 12.- 10 26.- 30 7.-

Exercise 4.8

The table contains the figures of monthly loans (in millions of CZK) given by a bank in
1997. Calculate the basic and chain indexes of the amounts loaned. For the base
period take (a) January, (b) July. What was the average index of monthly loans?

I II III IV V VI VII VIII IX X XI XII

53.2 56.2 49.5 48.0 47.6 52.8 54.5 42.9 49.2 56.0 57.1 55.0

Exercise 4.9

Use the basic index (with 1978 as the base period) and chain index to analyse the
data from Exercise 3.10 on the average fixed assets (in millions of CZK) for the years
1978 to 1985. On the assumption that the trend of the time series does not change

81
use the average chain index to make a forecast of the average fixed assets in 1987.

Questions

1. Define an index and describe its types.

2. How are single indexes determined and what are their properties?

3. How are group indexes determined and what are their properties?

4. How are composite indexes determined and what are their properties?

5. Define basic and chain indexes and show their use.

6. What are the drawbacks of the Laspeyres and Paasche index? Exemplify by a
particular market basket.

82
 5. MATHEMATICAL STATISTICS

Random sample and its characteristics

Using the methods of mathematical statistics, we can describe quantities of

random character from observed values. Most frequently, we try to establish the
properties of the probability distribution of a random variable - estimating parameters
or quantitative characteristics, testing the hypotheses that they have certain
properties, analysing the relationships between them etc.

If we conduct n experiments whose results are the values of a random

variable X with a distribution function F(x,-) where - is the parameter (or a vector
parameter or its function) of a given probability distribution, we, in fact, observe a
random vector X = (X1, ..., Xn) whose components are independent random variables
Xi which all have same distribution function equal to that of X. This random vector is
called a random sample or a simple random sample from random variable X. Its size
is n and its joint distribution function is
n
F(x, -) = F(x1, -) ... F(xn, -) = – F( xi , -) .
i 1

We define a random sample from a random vector in a similar way.

The numbers x1, ..., xa where xi is an observed value of Xi, i = 1,..., n are
said to be sample data of size n. Sample data processing is described in Chapter 2.

A function of a random selection T(X1, ..., Xn) is called a sample characteristic

or a statistic. The value t = T(x1, ..., xn) it takes on at sample data x1, ..., xn is said
to be an empirical characteristic or an observed value of statistic T. The following
statistics are frequently used:

1 n
a) sample mean X ¦ Xi ,
ni 1
1 n 2
b) sample variance S2 ¦
n i1
Xi  X ,

c) sample standard deviation S S2 ,

83
 1 n
¦ X i  X Yi  Y
ni1
d) sample correlation coefficient R .
SX SY
for a random sample from random vector (X, Y) where S(X) and S(Y) are the sample
standard deviations of X and Y.

We have:

DX n 1
a) EX E X, DX , E S2 DX .
n n
XP
b) If X has a normal distribution N(P,V2), then n  1 has the so-called
S
nS2
t-distribution S(n - 1) or Student's distribution and has the so-called
V2
chi-square distribution F2 n  1 .

You can find more information on statistics and their distributions in [4].

Parameter estimation

Point and interval estimations

We usually do not know the real value of a parameter - of the probability

distribution of an observed random variable X (or a random vector) and we try to
estimate it using sample data. These parameters often include number
characteristics of a random variable such as its expected value, variance, etc. We
can either make a point estimate or an interval estimate.

An estimator T of a parameter - is a statistic T(X1, ..., Xn) that assumes

values close to the parameter - whatever its value is. An estimator T is unbiased if its
expected value E(T) = -. If the variance of such an estimator is the least possible of
all unbiased estimators, we call T the minimum variance unbiased estimator. An
estimator T is called consistent, if lim P T  -  H 1 for an arbitrary real H ! 0 .
nof

84
Further types of estimators (such as maximum likelihood estimators) are described in
[1] and [4]. It holds:

a) X is an unbiased consistent estimator of the mean value E(X).

n
b) S2 is an unbiased consistent estimator of the variance D(X).
n 1
c) If X is normally distributed, the estimators under a) and b) are also the minimum
variance estimators .

A point estimate of a parameter - is the value t T x 1,...,x n an estimator T

assumes for sample data x1,...xn . Point estimates of the basic number
characteristics are calculated as follows:

n 2 n
EX x, D X s, VX s, U X, Y r .
n 1 n 1
An interval estimator for a parameter - with confidence 1 - D is a pair of
statistics T1;T2 such that
P T1 d - d T2 1 D

for any value of -. A 1 - D confidence interval or a 1 - D interval estimate for a

parameter - is an interval t1; t 2 where t1, t 2 are the values the statistics T1, T2

assume for a sample data.

The confidence 1 - D is mostly selected to be close to one, usually 0.95 or

0.99. Sometimes the confidence is shown as a percentage. Confidence intervals are
either one-sided or both-sided the former being limited on one side and the latter on
both sides. A confidence interval usually becomes shorter as n grows and longer as
the confidence increases. Given a confidence interval's length the size of sample
data needed can be established [1], [2] and [3].

Estimations of normal distribution parameters

We assume that the observed random variable X, or the random vector X, Y

is normally distributed with parameters P, V 2 , or U respectively. In the sequel we will

85
focus on both-sided interval estimations. One-sided estimations and interval
estimations for distributions other than normal can be found in [1].

Point estimators are calculated as follows

n 2 n
P x, V2 s, V s , U=r.
n 1 n 1

A (1- D) confidence interval for the mean value P where the variance V 2 is
unknown is given by the formula

s s
x t 1D 2 , x  t 1D 2 ,
n1 n 1

where t 1D 2 is the 1  D 2 - quantile of a t-distribution with n - 1 degrees of freedom

shown in Table T2.

A (1-D) confidence interval for the variance V 2 is given by

n s2 n s2
, ,
F 12D 2 F D2 2

where F P2 is the P-quantile of a chi-square distribution with n - 1 degrees of freedom

listed in Table T3. By taking the square root of this confidence interval we get an
confidence interval for the standard deviation V.

E x a m p l e 5.1

By measuring the lengths of 10 rollers sample data have been obtained with sample
characteristics x = 5.37 mm, s2 = 0.0019 mm2 and s = 0.044 mm (see Example
2.1). Calculate the minimum variance unbiased point estimates for the mean value,
variance and standard deviation. Assuming that the observed length is normally
distributed, calculate 0.95 confidence intervals for these characteristics.

S o l u t i o n:

The point estimates are calculated as follows:

mean value P = 5.37 mm ,

86
 10 2
variance V2 = 0.0019 = 0.00211 mm ,
9
standard deviation V= 0.00211 # 0.046 mm.

A 0.95 confidence interval for the mean value P is calculated for t0.975 = 2.262 for 9
degrees of freedom using Table T2,

0.0019 0.0019
P  <5.37  2.262; 5.37 + 2.262> # <5.337; 5.403> mm.
10  1 10  1

A 0.95 confidence interval for variance V2 is calculated for F02,025 = 2.70 and F 02.975 =

= 19.0 for 9 degrees of freedom using table T3,

10(0.0019) 10(0,0019)
V2  < ; > # <0.00100; 0.00704> mm2 ,
19.0 2.70

which means that a 0.95 confidence interval for standard deviation V is

V  < 0.00100 ; 0.00704 > # <0.0316; 0.0839> mm.

A (1-D) confidence interval for the correlation coefficient U for n t 10 is given

by

tgh z1 , tgh z 2 ,

where

u1D 2 u1D 2 1 § 1 r r · ez  e z e2 z  1
z1 w , z2 w , w ¨ ln  ¸ , tgh z
n3 n3 2 © 1  r n  1¹ ez  e z e2 z  1

and u1D 2 is the 1  D 2 - quantile of the standard normal distribution tabulated in

Table T1. For 1 - D = 0.95 and 1 - D = 0.99, we have u0.975 1.960 and u0.995 2.576

respectively.

E x a m p l e 5.2

A survey launched to examine the costs and price of a product manufactured by 10

different companies two-dimensional provided sample data to calculate a correlation

87
coefficient of r = 0.82482 (see Example 2.3). Calculate the minimum variance
unbiased point estimator and find a 0.99 confidence interval for the correlation
coefficient U of the parent population.

S o l u t i o n:

The point estimate calculated for the correlation coefficient is U = 0.82482.

Substituting it, we get

1 § 1  0.82482 0.82482 ·
w ¨ ln  ¸ # 1.21753 .
2 © 1  0.82482 10  1 ¹

In Table T1 we can find u0.995 = 2.576 so that

2.576 2.576
z1 1.21753  # 0.24397 , z 2 1.21753  # 2.19110
10  3 10  3

and a 0.99 confidence interval for the correlation coefficient U is shown below

U  tgh0.24397; tgh2.19110 # 0.239242; 0.975313 .

Estimations of parameter of a binomial distribution

Let a random variable X have an alternative distribution with the parameter

equal to p, that is, a binomial distribution Bi(1; p). When estimating p, we actually
estimate the ratio p of the parent population with a desired property where Xi
assumes values xi = 1 or 0 depending on the i-th element chosen at random having
or having not the property. Let x be the number of elements that have the property in
a random sample of size n. Thus we have
n
x= ¦x .
i 1
i

The minimum variance unbiased point estimator is

x
p= .
n

The interval estimator for n > 30 is

88
 x § x· x § x·
¨1  ¸ ¨1  ¸
x n © n¹ x n © n¹
 u1 D / 2 ;  u1 D / 2
n n n n

where u1D 2 is the 1  D 2 - quantile of the standard normal distribution N(0;1),

which can be found in Table T1.

E x a m p l e 5.3

When asked about a new product by a marketing research agency, 80 out of the 400
customers of the STAMET supermarket answered that they would buy it. Calculate
the minimum variance unbiased point estimator and find an D confidence interval for
the ratio p of such customers to all the STAMET customers.

S o l u t i o n:
80
Since x = 80 and n = 400, the point estimator assumes the value p = = 0.2, or
400
20% of the customers.

For confidence 0.95 we have u0.975 = 1.960, which yields the following 0.95
confidence interval for p

80 80 80 80
(1  ) (1  )
80 400 400 ; 80 + 1.960 400 400 ! =...=
p - 1.960
400 400 400 400

=  0.1608; 0.2392 !.

Similarly, a 0.99 confidence interval is calculated to be

p   0.1485; 0.2515 !.

We can say with a 0.95 or 0.99 confidence that about 16% to 24% or 15% to 25% of
the STAMET customers will buy the new product. If there are about 10 000 STAMET
customers it may be expected that 2 000 products will be sold. A 0.95 confidence
interval tells us that STAMET will sell approximately 10 000(0.16) = 1600 to
10 000(0.24) = 2400 new products.

89
 Testing statistical hypotheses

Statistical hypothesis and its testing

When observing random variables we must often test certain conditions or

assumptions on their properties using experiment data. For example we might want
to make a decision whether a parameter - has the value - o , writing H : - - o . If

H : - z - o , we call it a two-tailed alternative hypothesis and if H : - ² - o or - ¢ - o ,

we have a one-tailed alternative hypothesis.

To test a hypothesis H : - - o a suitable statistic T X 1 , ... , X n is constructed

the so-called test criterion or test statistic. For - - o , the range of values of the test
criterion is divided into two disjunct subsets - the critical range WD and its

complement W D in such a way that, for - - o , the probability of T X 1 , ... , X n

taking on a value from W D is 1 - D . The number D ! 0 is called a level of significance

or level of significance and is chosen close to zero, usually 0.05 or 0.01.

If, for sample data x1 , ... , xn , the test criterion assumes a value t =

= T x1 , ... , xn from the critical range that is t  WD , we reject the hypothesis H and

do not reject the hypothesis H. If, on the other hand, t lies outside the critical range
or t  W D , we reject the hypothesis H and do not reject the hypothesis H at the level

of significance D. If a hypothesis H or H is rejected, it does not necessarily mean

that its validity has been proved since via the random sample we have only acquired
information which does not suffice to reject it. Whenever possible, the size of the
sample data should be increased before accepting a given hypothesis H and it
should be tested again.

When testing a hypothesis H, the following errors may be made

(1) the so called first type error, when H is true but t  WD , so that we reject it (the

probability of such an error is P T  WD H D ),

90
(2) the so called second type error, when H is not true but t  WD , so that we do not

reject it (the probability of this error is P T  WD H ).

Since a test criterion T is a random variable, the range W D is often in the form

of an interval t 1 ; t 2 where t 1 , t 2 are quantiles of the test statistic T (the so-called

critical values) as in confidence intervals. You can find more about statistical
hypotheses and their testing in [1], [2], [3] and [4].

Testing hypotheses concerning parameters of normal distribution

We assume that random variables and vectors are normally distributed. The
following testing criteria are for two-tailed alternative hypotheses such as H : P z P0
except for the variance equality test.

Assumptions on distribution types may be tested using the so-called goodness

of fit tests (such as Pearson's test or Kolmogorov's test) or we can use a rough
graphical test using the so-called probability paper or its graphical version on a
computer. These tests and further details on testing hypotheses with one-tailed
alternative hypotheses tests about parameters of other distributions and other tests
can be found in [1], [2] and [4].

Testing a hypothesis H : P P0 with unknown variance V 2 . The test criterion

is calculated using the formula
x  P0
t n 1,
s

and W D  t 1D 2 ; t 1D 2 where t 1D 2 is the 1  D 2 -quantile of the t-distribution

with n - 1 degrees of freedom. These values can be found in Table T2. This is the
so-called one sample t - test.

91
E x a m p l e 5.4

By measuring the lengths of 10 rollers, empirical characteristics x = 5.37 mm and

s2 = 0.0019 mm2 have been established (see Problem 2.1). At the 0.05 level of
significance test the hypothesis that the mean value of the roller length measured is
5.40 mm, and so H : P = 5.40.

S o l u t i o n:

The test criterion assumes the value

5.37  5.40
t 10  1 # -2.0647.
0.0019

For 10 - 1 = 9 degrees of freedom, we find t0.975 = 2.262 in Table T2 so that W0.05 =

= <-2.262; 2.262>. Since t  W0,05 , we do not reject the hypothesis. To test this

hypothesis we could also use the 0.95 confidence interval from Example 5.1. Since
this interval includes the hypothetical value of 5.40, we do not reject the hypothesis at
the level of significance 1 - 0.95 = 0.05.

Testing hypothesis H : V2 V02 . The test criterion is given by

n s2
t
V02

and W D F 2D 2 ;F12D 2 where FP2 is a P-quantile of the chi-squared distribution with

n - 1 degrees of freedom listed in Table T3.

E x a m p l e 5.5

At the 0.05 level of significance test a hypothesis that the variance of the value of the
roller length measured in Example 5.2 is 0.0025 mm2 so that H : V2 = 0.0025.

S o l u t i o n:

The test criterion assumes the value

10(0.0019)
t = 7.6 .
0.0025

92
For 10 - 1 = 9 degrees of freedom we determine F 02.025 = 2.70 and F 02.975 = 19.0 from

Table T3 so that W0.05 = <2.70; 19.0>. Since t  W0.05 , we do not reject the

hypothesis.

Testing hypothesis H : U U0 . The value of the test criterion is calculated for

n t 10, r z 1 and U 0 z 1 using the formula

§ 1 r 1  U0 U · n3
t ¨¨ ln  ln  0 ¸¸
© 1 r 1  U0 n  1 ¹ 2

and W D  u1D 2 ; u1D 2 where u1D 2 is the 1  D 2 - quantile of the normal

distribution N(0, 1), which can be found in Table T1.

E x a m p l e 5.6

By monitoring the costs X and prices Y of an identical product with ten

manufacturers, a two-dimensional statistical data have been collected and a sample
correlation coefficient r = 0.82482 calculated (see Example 2.3). At the 0.01 level of
significance, test the hypothesis that the random variables X and Y are not correlated
(independent with respect to normal distribution) so that H : U = 0.

S o l u t i o n:

The test criterion assumes the value

§ 1  0.82482 1 0 0 · 10  3
t ¨ ln  ln  ¸ # 3.1001.
© 1  0.82482 1  0 10  1 ¹ 2

For the given level of significance we find u0.995 = 2.576 in Table T1 so that W0.01 =

= <2.576; 2.576 >. Since t  W0.01 , we reject the hypothesis and consider X, Y as

dependent.

Testing hypothesis H : P X PY for pairs. Denote by x i , yi , where

i = 1,…,n, the values of pairs observed in the random vector (X, Y), by di xi  y i

their differences and by d and s 2 d their empirical characteristics . The test criterion
is given by

93
 d
t n1
sd

and W D  t 1D 2 ; t 1D 2 , where t 1D 2 is the 1  D 2 - quantile of the chi-squared

distribution with n -1 degrees of freedom, which can be found in Table T2. This is the
so-called t - test for paired values.

Example 5.7

Using two thermometers the following pairs of temperature values have been
measured over eight days: (xi; yi) = (51.8; 49.5), (54.9; 53.3), (52.2; 50.6),

(53.3; 52.0), (51.6; 46.8), (54.1; 50.5), (54.2; 52.1), (53.3; 53.0) (oC). At the level of
significance 1%, test the hypothesis that the difference of the mean values is
insignificant so that H : P(X) = P(Y).

S o l u t i o n:
For di = xi - yi, i = 1,...,8, we get d = 2.2 oC and s(d) = 1.3172 oC. The test criterion
assumes the value

2.2
t= 8  1 # 4.4190.
1.3172

For 8 - 1 = 7 degrees of freedom we have t0.995 = 3.499 from Table T2 so that W0.01 =

= <-3.499; 3.499>. Since t  W0.01 , we reject the hypothesis at the level of

significance 1%. The difference between the two measurements is statistically

significant.

For the following tests we assume that by observing two independent random
variables X and Y normally distributed with parameters P X , V2 X and P Y , V2 Y
sample data of sizes n1 and n 2 have been obtained.

Testing hypothesis H : P X  P Y P0 with unknown variances. The test

criterion is calculated as follows

94
 x  y  P0 n1n2 n1  n2  2
t
2
n1 s x  n2 s y 2 n1  n2

and W D  t 1D 2 ; t 1D 2 where t 1D 2 is the 1  D 2 - quantile of the chi-squared

distribution with n1  n 2  2 degrees of freedom listed in Table T2. This is the so-
called two-sample t - test.

E x a m p l e 5.8

By testing the strength of wire manufactured by two different technologies

two sample data have been obtained with the following sample
characteristics n1 = 33, x = 5.4637 kN, s2(x) = 0.3302 kN2, n2 = 28, y = 6.1179 kN,

s2(y) = 0.4522 kN2. At the level of significance 0.05 test the hypothesis that the
different technologies do not affect the expected value of the wire strength (assuming
that the variances V2 (X) and V2 (Y) are the same) so that H : P(X) - P(Y) = 0.

S o l u t i o n:
The test criterion assumes the value

5.4637  6.1179  0 33(28) 33  28  2

t # 4.030.
33(0.3302)  28(0.4522) 33  28

For 33 + 28 - 2 = 59 degrees of freedom we get t0.975 = 2.001 by interpolating in Table

T2 so that W0.05 = <-2.001; 2.001>. Since t  W0.05 , we reject the hypothesis. The

different technologies do affect the expected value of the wire strength.

Testing hypothesis H : P X  P Y P0 for unknown variances V 2 X z V 2 Y .

The test criterion is calculated as
x  y  P0
t
s2 x s2 y

n1  1 n2  1

and W D  t1 D / 2 ; t1 D / 2 where

95
 s2 ( x ) s2 ( y )
t( x )  t( y )
n1  1 n2
t1 D / 2
s2 ( x ) s2 ( y )

n1  1 n2  1

and t(x) or t(y), is the 1  D 2 - quantile of the chi-squared distribution with n1 – 1 or

n2 – 1 degrees of freedom respectively shown in Table T2. This is the so-called two-
sample t - test.

E x a m p l e 5.9

Surveys made to determine the mean service life of products in two different systems
of extreme conditions yielded two sets of sample data with the following sample
characteristics n1 = 21, x = 3.581, s2(x) = 0.114, n2 = 23, y = 3.974, s2(y) = 0.041 (the
length of the life is in hours). At the level of significance 0.05, test the hypothesis that
the second system of extreme conditions increases the mean service life by 0.5 h as
compared with the first one (assuming different variances V2(X) and V2(Y)) so that
H : P(X) - P(Y) = - 0.5.

S o l u t i o n:

The test criterion assumes the value

3.581  3.974  ( 0.5)

t # 1.2303.
0.114 0.041

21  1 23 2  1

Entering Table T2 at 1- D/2 = 0.975 we see that t(x) = 2.086 for 21 - 1 = 20 degrees
of freedom and t(y) = 2.074 for 23 - 1 = 22 degrees of freedom. Then we can
calculate

0.114 0.041
2.086  2.074
t 0,975 21  1 23  1 # 2.083.
0.114 0.041

21  1 23  1

and W0.05 = <-2.083; 2.083>. Since t  W0.05 , we do not reject the hypothesis that

the second system increases the mean service life by 0.5 h.

96
 Testing hypothesis H : V2 X V2 Y against alternative hypothesis

H : V2 X ! V2 Y . The test criterion is calculated as follows

n1s2 ( x )
n1  1
t t1
n2 s 2 ( y )
n2  1

and WD 1; F1 D where F1D is the ( 1  D ) - quantile of the Fisher - Snedecor

distribution with k1 n1  1 and k 2 n2  1 degrees of freedom listed in table T4. If

t  1 we swap X and Y. This is the so-called F - test or Fisher's test. It is used to test
assumptions on variances made in the previous tests.

E x a m p l e 5.10

At the level of significance 0.05, test the hypothesis that the variances V2 (X) > V2 (Y)
are different in Example 5.9 where s2(x) = 0.114, n1 = 21, s2(y) = 0.041, n2 = 23.

S o l u t i o n:

Let us, on the contrary, test the hypothesis H : V2 (X) = V2 (Y).

The test criterion assumes the value

21(0.114)
t 21  1 # 2.7926.
23(0.041)
23  1

For k1 = 21 - 1 = 20 and k2 = 23 - 1 = 22 degrees of freedom we obtain F0.95 = 2.071

by interpolating in Table T4 so that W0.05 = <1; 2.071>. Since t  W0.05 , we reject the

hypothesis. We consider the assumption that the variances are different as correct.

Testing hypotheses on the parameter of a binomial distribution

Let us observe a random variable X which has an alternative probability

distribution with a parameter p, thus it has the binomial distribution Bi(1; p). When
testing a hypothesis H : p = p0 we actually test a hypothesis that the ratio of those
elements of the parent population that have a desired property is p0 on the basis of

97
the fact that x elements out of the n elements in a random sample have the property
(see estimations of parameters).

Testing hypothesis H : p = p0 . The testing criterion is calculated for n ! 30 as

follows
x
 p0
t n
p0 (1  p0 )
n

and W D  u1D 2 ; u1D 2 where u1D 2 is the 1  D 2 - quantile of the normal

distribution N(0, 1) which can be found in Table T1.

E x a m p l e 5.11

According to an expert's opinion 20% of the customers will be interested in a new

product. Out of the 400 customers asked, 62 showed interest. At the 0.05
significance level test the hypothesis that the expert's forecast is real, that is H : p =
= 0.2.

S o l u t i o n:

For x = 62 and n = 400 the testing criterion assumes the value

62
 0.2
400  0.045
t= = -2.25 .
(0.2)(1  0.2) 0.02
400

From Table T1 we get u0.975 = 1.960. Since t = -2.25  W0,05 = <-1.960; 1.960>, we

reject the hypothesis that 20% of customers will be interested at the 0.05 significance
level. The real interest will probably be lower. Note that we would not reject the
hypothesis at the 0.01 significance level since u0.995 = 2.576.

For the following test we assume that we observe two independent random
variables X, Y which have s alternative distributions with parameters p1, p2 and that
two independent sample data have been obtained of sizes n1 , n 2 respectively and

98
the respective numbers x, y of elements with the desired property (see estimations
of parameters).

Testing hypothesis H : p1 = p2. The testing criterion can be calculated as

shown below on the assumption that n1 ! 50 a n 2 ! 50

x y

n1 n 2 n1n 2
t ,
f (1  f ) n1  n 2
xy
for f and W D  u1D 2 ; u1D 2 where u1D 2 is the 1  D 2 - quantile of the
n1  n2
normal distribution N(0, 1) with values shown in Table T1.

E x a m p l e 5.12

Shop inspectors bought 250 items of food and 200 items of hard goods to test their
quality. Subsequently, they found 108 items of food and 73 items of hard goods to be
defective. At the 0.05 significance level test, if the quality of food and hard goods is
equal, that is the hypothesis H: p1 = p2 where p1, p2 are the theoretical ratios
(probabilities) of buying defective items for the given kinds of goods.

S o l u t i o n:

For x = 108, n1 = 250, y = 73, n2 = 200 we get

108  73
f # 0.40222
250  200

and the testing criterion assumes the value

108 73

250 200 250(200) 0.067(10.5409)
t # # 1.4403 .
0.40222(1  0.40222) 250  200 0.49035

We establish u0.975 = 1.960 from Table T1. Since t = 1.4403  W0.05 = <1.960;

1.960>, at the 0.05 significance level we do not reject the hypothesis that the
probabilities of buying a defective item are the same for both kinds of goods and
consider both kinds of goods to be of equally bad quality.

99
 Regression analysis

Regression function

An important role of statistics is to find and analyse dependencies of variables

whose values we observe when conducting experiments. Because of their random
character, the independent variables are represented by a random vector
X X 1 ,..., X k and the dependent variable is represented by a random variable Y.
We employ regression analysis to determine the dependence of Y on X and this
dependence is expressed by a regression function

y M x, ȕ E Y/X x ,

where x = x1 ,..., x k is a vector of independent variables (the value of X), y is a

dependent variable (the value of Y) and E E 1 ,..., E m is a vector of parameters, the

so-called regression coefficients. X can also be a non-random vector as it is often the

case in applications, for example if the variances of the vector constituents
Xj , j 1, ... , m , can be neglected as compared to the variance of Y.

When determining the dependence of Y on X we conduct experiments to

obtain a sequence of points >x 1,y1 @ ,..., >x n,y n @ where y i is the result of measuring Yi
and x i is the resulting value of measuring the vector of independent variables, i =
= 1,..., n. To obtain estimates of the regression coefficients E j , j = 1,..., m, we

minimize the so-called residual sum of squares

n

¦ >y @
2
S* i  M xi ,E
i 1

using the least-square method.

Linear regression function

A linear regression function (linear with respect to the regression coefficients)

can be written in the form

100
 m
y ¦E f
j 1
j j x ,

where fj x are known functions not containing E 1 ,..., E m .

Let us consider a model based on the following assumptions:

1. The vector x is non-random, so that the functions fj x assume non-random

values fji fj x i for j = 1,..., m and i = 1,..., n.

2. The matrix F fji is an (m, n)-matrix with rank m < n.

m
3. For i = 1,..., n, random variable Yi has the expected value E Yi / Xi xi ¦E f
j 1
j ji

and a constant variance D Yi / X i xi V2 ! 0 .

4. Random variables Yi are not correlated and are normally distributed for i = 1,..., n.

Using the following denotations

§ b1 · § y1 · § f1( x ) ·
¨ ¸ ¨ ¸ ¨ ¸ T
b ¨ # ¸, y ¨ # ¸ , f(x) ¨ # ¸ , H FF , g = Fy ,
¨b ¸ ¨y ¸ ¨ f (x) ¸
© m¹ © n¹ ©m ¹
we get:

a) For j = 1,..., m, E j is the minimum variance unbiased point estimate for the

regression coefficient b j where b is the solution of the system of linear algebraic

equations (the so-called system of normal equations)

Hb = g.

b) The minimum variance unbiased point estimate for the linear regression function
is
m
y ¦b f
j 1
j j x .

c) The minimum variance unbiased point estimate for variance V 2 is

*
S min
s2 ,
nm

101
 where
2
n § m · n m
S *
min ¦ ¨ y i  ¦ b j f ji ¸
¨
i 1©
¸ ¦ y  ¦b g
2
i j j
j 1 ¹ i 1 j 1

and g j is an element of matrix g.

d) The (1 - D) - confidence interval for the regression coefficient E j is

bj  s h jj t 1 D 2 ; b j  s h jj t 1 D 2 ,

j = 1, ..., m, where h jj is the jth diagonal element of the matrix H1 and t 1D 2 is

the 1  D 2 - quantile of the tdistribution with n - m degrees of freedom.

e) The (1 - D) - confidence interval for the expected value of y is

m m

¦ b jfj (x )  t1D / 2s h* ; ¦ b jfj (x )  t1D / 2s h* ,

j 1 j 1

where h* = f(x)TH-1f(x) and t 1D 2 is the 1  D 2 - quantile of the t-distribution with

n - m degrees of freedom. A (1 - D) - confidence interval for an individual

function value of y is calculated in a similar way taking 1 + h* instead of h* .

f) To test the hypothesis H : Ej E j0 with an D level of significance where j is an

arbitrary index, j = 1,..., m, the testing criterion must be calculated

b j  E j0
t
s h jj

and WD  t 1D 2 ; t 1D 2 , where t 1D 2 is the 1  D 2 - quantile of the

t  distribution with n - m degrees of freedom.

The simplest and the most used linear regression function is the so-called
regression line
y E1  E2 x .

For this function, k = 1, x = x1 = x (we write x instead of x1), m = 2, f1(x) = 1, f2(x) = x,

so that

102
 § y1 ·
§1 " 1 · ¨ ¸
F ¨¨ ¸¸ , y ¨ # ¸.
© x1 " x n ¹ ¨y ¸
© n¹
When calculating with a pocket calculator, we can use the following explicit formulas
n
( ¦ stands for ¦ i 1
):

§ ¦1
1) H ¨
¦x i ·
¸ , g
§ ¦ yi ·
¨ ¸ ,
¨¦ x 2¸ ¨¦ x y ¸ ¦1 n ,
© i ¦x i ¹ © i i¹

2 n¦ x i y i  ¦ x i ¦ y i
2) det H n¦ x i2  ¦x i , b2 , b1 y  b2 x ,
det H
*
2 Smin
3) S *
min ¦ yi  b1  b2 xi ¦y 2
i  b1 ¦ yi  b2 ¦ xi yi , s 2

n2
,

4) h 11 ¦x i
, h22
n
,
det H det H
2 2
* 1 xx 1 n xx
5) h  2
 ,
n ¦ x i2  n x n det H

*
Smin
6) r 1 2
r( x, y ) , where r(x,y) is the correlation coefficient – see
¦y 2
i n y

Chapter 2.

Example 5.13

A survey at eight tax consultancy companies chosen at random produced the

following numbers of employees x and annual sales y (in millions of CZK):

xi 3 5 5 8 9 11 12 15

yi 0.8 1.2 1.5 1.9 1.8 2.4 2.5 3.1

Find out how the annual sales of a company depend on the number of its employees.
Using the formula y = E1 + E2 x, calculate a 0.95 confidence interval for E2 , with a
0.05 level of significance test the hypothesis H : E1 = 0.2, find the minimum variance
unbiased point estimate and a 0.95 confidence interval for y(10). Using a graph and
the correlation coefficient analyse the suitability of the regression function.

103
S o l u t i o n:

The table below contains auxiliary calculations:

i xi yi xi2 xiyi yi2

1 3 0.8 9 2.4 0.64

2 5 1.2 25 6.0 1.44

3 5 1.5 25 7.5 2.25

4 8 1.9 64 15.2 3.61

5 9 1.8 81 16.2 3.24

6 11 2.4 121 26.4 5.76

7 12 2.5 144 30.0 6.25

8 15 3.1 225 46.5 9.61

6 68 15.2 694 150.2 32.80

This is a regression line and so using the above formulas and table, for n = 8,
§ 8 68 ·
we can set up the matrix H = ¨¨ ¸¸ with det H = (8)(694) - 682 = 928. This
© 68 694 ¹
yields the point estimate for E2 :

(8)(150.2)  (68)(15.2)
b2 = = 0.1810344 # 0.181.
928

Next we have x = 68/8 = 8.5 and y = 15.2/8 = 1.9, so that the point estimate for E1
is
b1 = 1.9 – (0.1810344)(8.5) = 0.3612068 # 0.361.

Thus we get the point estimate for the regression function y = 0.361 + 0.181x.

The minimum value of the residual sum of squares is

Smin = 32.80 – (0.3612068)(15.2) – (0.1810344)(150.2) # 0.1182758

and the point estimates for V2 and V, are the following

s2 = 0.1182758/(8-2) = 0.0197126, resp. s = 0.0197126 # 0.1404017.

104
The diagonal elements of H-1 are h11 = 694/928 # 0.7478448 and h22 = 8/928 #
# 0.00862069. In Table T2, for 8 - 2 = 6 degrees of freedom, we find t0.975 = 2.447.
The 0.95 confidence interval for the regression coefficient E2 is

E2  <0.1810344 – (2.447)(0.1404017) 0.00862069 ;

0.1810344 + (2.447)(0.1404017) 0.00862069 > =

= <0.1491353; 0.2129334> # <0.149; 0.213>.

We have found a point estimate of 181 000 CZK for the increase in annual sales
corresponding to an increase in the number of employees by one. A 0.95 confidence
interval for this value is 149 000 CZK to 213 000 CZK.

For H : E1 = 0.2 the testing criterion assumes the value

0.3612068  0.2
t= # 1.3277.
0.1404017 0.7478448

For the alternative hypothesis H : E1 z 0.2 we get W0,05 = <-2.447; 2.447>. Since

t  W0.05 , we do not reject the hypothesis E1 = 0.2 at a 0.05 level of significance. As

it is, at this level of significance we would not actually reject the hypothesis that a
company without employees (the owners themselves work), since y(0) = E1 , will
have annual sales of about 200 000 CZK.

The point estimate for the average and individual sales at a company that has
ten employees is the following

y(10) = 0.3612068 + (0.1810344)(10) = 2.1715508 # 2.172.

For the given company annual sales of about 2 172 000 CZK may be expected Since

1 8(10  8.5)2
h* =  = 0.1443965,
8 928

is a 0.95 confidence interval for the annual sales of a company with ten employees,

y(10)  <2.1715508 – (2.447)(0.1404017) 0.1443965 ;

2.1715508 + (2.447)(0.1404017) 0.1443965 > =

= <2.0409985; 2.3021031> # <2.040; 2.302>.

105
With probability 0.95 it may be expected that the average annual sales of such a
company will range between 2 040 000 CZK and 2 302 000 CZK.

If we employ 1 + h* instead of h* for the calculation, we get a 0.95 confidence

interval for the value of the annual sales of a firm with ten employees

y(10)  <2.1715508 – (2.447)(0.1404017) 1.1443965 ;

2.1715508 + (2.447)(0.1404017) 1.1443965 > # <1.804; 2.539>.

With a 0.95 probability it may be expected that the annual sales (its individual value)
of such a company will range from 1 804 000 CZK to 2 539 000 CZK.

The correlation coefficient (calculated by the formula in Chapter 2) is r =

= 0.984798. It follows from the graph in Figure 5.1 and from the value of the
correlation coefficient that the form of the regression function chosen fits the given
dependence quite well.

Dependence of annual sale on number of employees

3.6
3.2
Annual sale (million of CZK)

2.8
2.4
2.0
1.6
1.2
0.8
0.4
0.0
0 3 6 9 12 15 18
Number of employees

Fig. 5.1

The regression functions are either linear or nonlinear (with respect to their
regression coefficients). Some of the nonlinear regression functions can be

106
transformed into linear ones using a suitable linearization (such as taking a logarithm
of a power or an exponential function). You will find more details about linearization,
tests of the suitability of a linear regression function, regression diagnostics and other
topics in [1], [2], [3] and [4].

Exercises

E x e r c i s e 5.14

Find the minimum variance unbiased point estimate and a 0.99 confidence interval
for the parameters P and V2 of a normal distribution, using sample data from a
random sample of size n = 18 with the sample mean x = 50.1 and variance
s2 = 17.64.

R e s u l t: P = 50.1; V2 # 18.678; P  <47.09; 53.10>; V2  <8.894; 55.705>

E x e r c i s e 5.15

Sample data of size n = 12 have sample mean x = 77.55 and variance s2 = 1045.65.
Calculate the point estimator and find a 0.99 confidence interval for P and V of the
parent population.

R e s u l t: P = 77.55; V # 33.78; P  <47.267; 107.833>; V  <21.638; 69.47>

E x e r c i s e 5.16

A total of 100 workers of the same category selected at random have been asked
about their wages per hour and the empirical characteristics x = 28.64 CZK and s2 =
= 1.1979 CZK have been calculated. Find the minimum variance unbiased point
estimate and a 99% confidence interval for the expected value of wages per hour P
and standard deviation V provided that the parent population is normally distributed.

R e s u l t: P = 28.64 CZK; V # 1.10 CZK;

P  <28.35; 28.93> CZK; V  <0.93; 1.34> CZK

107
E x e r c i s e 5.17

From fifteen independent measurements of the maximum velocity of an aeroplane

the minimum variance unbiased point estimates 424.7 ms-1 for the mean value and
8.7 ms-1 for the standard deviation have been calculated. Find a 95% confidence
interval for the expected value and standard deviation of the maximum speed
provided that it is normally distributed.

R e s u l t: P  <419.88; 429.52> ms-1 ; V  <6.37; 13.72> ms-1

E x e r c i s e 5.18

A total of fifty values of a random variable with a normal distribution N(P,V2) has been
used to calculate a sample mean x = 610 and a variance s2 = 2770.4. Find interval
estimates for P with confidences of 0.9; 0.95; and 0.99.

R e s u l t: <597.45; 622.55>; <594.97; 625.07>; <592.77; 627.33>

E x e r c i s e 5.19

Five independent and equally accurate measurements have been carried out to
determine the volume of a vessel: 4.781; 4.792; 4.795; 4.779; 4.769 (in litres). Find a
0.99 level of significance confidence interval for the expected value of the volume of
vessel provided that it is normally distributed.

R e s u l t: <4.761; 4.805> l

E x e r c i s e 5.20

A sample of size n = 128 has been used to determine the correlation coefficient and
the result was r = 0.560. Calculate the minimum variance unbiased point estimator
and find a 95% confidence interval for U.

R e s u l t: U = 0.560; U  <0.441; 0.679>

E x e r c i s e 5.21

In a total of 46 households chosen at random a survey has been made to find the
relationship between income X and expenses Y. A sample correlation coefficient of

108
r = 0.638 was calculated. Calculate the minimum variance unbiased point estimator
and determine a 0.95 confidence interval for the correlation coefficient provided that
income and expenses have a two-dimensional normal distribution.

R e s u l t: U = 0.638; U  <0.423 ;0.780>

E x e r c i s e 5.22

Questionnaires have been sent to 190 customers of a pension fund chosen at

random with questions about their annual income X, premium amount Y and
employment time Z. Calculations have been performed to determine the correlation
coefficients with the following results: U(X,Y) = 0.55; U(X,Z) = 0.30; U(Y,Z) = 0.37.
Find a 99% level of significance confidence intervals of the above coefficients
assuming a normal distribution.

R e s u l t: U(X,Y)  <0.42; 0.68>; U(X,Z)  <0.13; 0.47>; U(Y,Z)  <0.21; 0.53>

E x e r c i s e 5.23

During a check of expiry periods of a certain type of tinned meat in a food industry
warehouse, a total of 320 tins have been chosen at random. It has been found that in
59 of them the guarantee periods have expired. Calculate the minimum variance
unbiased point estimate and find a 95% level of significance confidence interval for
the percentage of tins in the warehouse with expired guarantee periods. Do the same
for a warehouse with 20 000 tins.

R e s u l t: p = 59/320 # 0.184 = 18.4 %; p  <0.142; 0.226> = <14.2; 22.6> %;

N = 3680; N  <2840; 4520>

E x e r c i s e 5.24

Within a random sample from tyres produced by a large European multi-national

company, 10% of the tyres do not comply with the rules of a new standard. Find a
95% confidence interval for the percentage p (of the whole parent population) of tyres
that do not comply with the new standard if the size of the random sample is a) n =
= 100, b) n = 400, c) n = 1600.

109
R e s u l t: <0.041; 0.159>; <0.071; 0.129>; <0.085; 0.115>

E x e r c i s e 5.25

Sample data of size n = 10 provide a sample mean of x = 32 and a variance of

s2 = 15. At a 0.05 level of significance test the hypothesis that the expected value of
the parent population is P = 30.

R e s u l t: t # 1.549; t0.975 = 2.262; we do not reject the hypothesis

E x e r c i s e 5.26

The following sample data have been provided by a random sample from normal
distribution.

xj* -2 -1 0 1 2 3

fj 1 4 7 3 3 2

At a 0.05 level of significance test the hypothesis P = 0.1.

R e s u l t: x = 0.45; s = 1.3592; t # 1.1224; t0.975 = 2.093; we do not reject the

hypothesis

E x e r c i s e 5.27

The expected value of humidity in roasted coffee is set at 4.2% and the standard
deviation at 0.4%. The actual percentages of humidity determined by analyzing 20
samples are the following: 4.5; 4.3; 4.1; 4.9; 4.6; 3.2; 4.4; 5.1; 4.8; 4.0; 3.7; 4.4; 3.9;
4.1; 4.2; 4.1; 4.7; 4.3; 4.2; 4.4. At a 5% level of significance test the hypotheses that
a) the expected value of humidity for the parent population complies with the
standard and
b) the standard deviation of humidity for the parent population complies with
the standard.

R e s u l t: a) t # 1.033; t0.975 = 2.093; we do not reject the hypothesis

b) t = 22.25; F 02.025 = 8.91; F 02.975 = 32.9; we do not reject the hypothesis

110
E x e r c i s e 5.28

Using sample data of size n = 10 and a variance of s2 = 2.0 at a 0.01 level of

significance, test the hypothesis that the variance of the parent population is V2 = 0.2.

R e s u l t: t = 100; F 02.005 = 1.73; F 02.995 = 23.6; we reject the hypothesis

E x e r c i s e 5.29

Using a random sample from a two-dimensional normal distribution sample data have
been obtained of size n = 44 and the correlation coefficient has been found to be
r = 0.7417. At a 1% level of significance test the hypothesis that the random variables
for the parent population are independent.

R e s u l t: H : U = 0; t # 6.110; u0.995 = 2.576; we reject the hypothesis

E x e r c i s e 5.30

With two-dimensional sample data of size n = 20 a sample correlation coefficient of

r = 0.3998 has been calculated. At a 5% level of significance test the hypothesis that
the random variables for the parent population are independent.

R e s u l t: H : U = 0; t # 1.747; u0.975 = 1.960; we do not reject the hypothesis

E x e r c i s e 5.31

To compare the accuracy of two methods of measurement a total of 8 measurements

have been carried out and the differences in the pairs of the corresponding
measurements have been calculated. In this way the average deviation d = 0.244
and standard deviation s(d) = 0.192 have been found. At a 0.05 level of significance
test the hypothesis that both methods may be taken for equally accurate.

R e s u l t: t # 3.362; t0.975 = 2.365; we reject the hypothesis

E x e r c i s e 5.32

Using two analytical scales a total of 10 samples have been weighed with the
following results: (xi; yi) = (25; 28), (30; 31), (28; 26), (50; 52), (20; 24), (40; 36),
(32; 33), (36; 35), (42; 45), (38; 40) (mg). With a 0.01 level of significance find out

111
whether the different results are statistically insignificant provided that they are
normally distributed.

R e s u l t: t # -1.13; t0.995 = 3.250; we do not reject the hypothesis, the different

results are statistically insignificant

E x e r c i s e 5.33

The following are the results of measurements made before and after a scale of a
packing machine has been calibrated: n1 = 12, x = 31.2 g, s2(x) = 0.770 g2 and
n2 = 18, y = 29.2 g, s2(y) = 0.378 g2. Suppose the variances are equal and the
distribution is normal. At a 0.05 level of significance test the hypothesis that the
expected value has not been changed by the calibration.

R e s u l t: t # 7.1; t0.975 = 2.048; we reject the hypothesis

E x e r c i s e 5.34

The average grading of a total of twenty study groups of students in a particular year
are shown in the table below:

xj* 1.70 1.86 2.01 2.23 2.27 2.411

fj 2 3 5 7 2 1

The overall average grading in the previous year for 20 study groups was y = 2.201
and the variance was s2(y) = 0.012 . Test the hypothesis that the average gradings of
the two years do not differ if we assume that the distribution is normal and variances
are equal.

R e s u l t: x = 2.0795; s2(x) = 0.0399; t # - 2.325;

t0.975 = 2.023 - we reject the hypothesis
t0.995 = 2.712 - we do not reject the hypothesis

E x e r c i s e 5.35

Two types of rope have been tested for tensile strength. Two samples of an equal
size of n = 18 have been taken and the following values have been calculated:

112
x = 3389.3 N, s2(x) = 1144.4 N2, y = 3339.2 N, s2(y) = 3453.5 N2. Assuming the
variances to be different test the hypothesis that the expected tensile strengths of the
ropes are the same. Use a 0.05 level of significance.

R e s u l t: t # 3.046; t 0.975 = 2.110; we reject the hypothesis

E x e r c i s e 5.36

Two sample data of sizes n1 = 20 and n2 = 10 with the following characteristics

x = 10.24; y = 11.09; s2(x) = 4.231 and s2(y) = 18.457 have been calculated using
two normally distributed random samples from random variables X and Y with
different variances. At a 1% level of significance test the hypothesis that the expected
values of X and Y are the same.

R e s u l t: t # -0.5637; t 0.975 = 3.212; we do not reject the hypothesis

E x e r c i s e 5.37

Two different methods have been used to determine the fat content of milk. Using the
first method for a sample of 12 analyses a variance of s2(x) = 0.0224 has been
calculated and with the second method a sample of 8 analyses has been used to
produce a variance of s2(y) = 0.0263. At a 0.01 level of significance test the
hypothesis that both methods are equally accurate in terms of variance.

R e s u l t: t # 1.23; F0.95 = 4.907; we do not reject the hypothesis

E x e r c i s e 5.38

Test the hypothesis that the variances in Exercise 5.19 are equal. Use a 0.05 level of
significance.

R e s u l t: t = 2.1; F0.95 = 2.41; we do not reject the hypothesis

E x e r c i s e 5.39

The Board of Directors of a large company consider selling shares of the company's
stock to their own employees. The estimate that about 20% of the employees will buy
the shares. A total of 400 employees chosen at random have been asked whether

113
they will buy the shares. The answer has been yes in 66 cases. Using a 0.05 level of
significance test the hypothesis that the directors' estimate is realistic.

R e s u l t: t = -1.75; u0.975 = 1.960; we do not reject the hypothesis, the estimate is

realistic

E x e r c i s e 5.40

A sample of size n = 200 has been taken from products manufactured using a new
technology. Out of those 200 products 31 have been found to be defective. Ascertain
that the new technology has changed the wastage rate of the products as compared
to a previous rate of 10% determined by long experience. Use a 1% level of
significance.

R e s u l t: t # 2.593; u0.995 = 2.576; we do not reject the hypothesis (the new

technology has changed the wastage rate)

E x e r c i s e 5.40

In two plants the same type of product is manufactured. The wastage rates in the two
plants should be the same as both use the same technology. In the first plant, 10
products out of a total of 200 chosen at random and checked are defective while in
the second plant it is 23 defective products out of a total of 250. Using a 0.01 level of
significance test whether there is a statistically significant difference in quality
between the two plants.

R e s u l t: t # -1.699; u0.995 = 2.576; we do not reject the hypothesis, (no statistically

significant difference exists between the two plants)

E x e r c i s e 5.41

A survey in a certain region has ascertained that out of a sample of 58 farmers 23

were ill while in a sample of 43 workers 28 were ill. At a 5% significance level test the
hypothesis that the sickness rate in workers is higher than in farmers.

R e s u l t: t # -2.534; u0.975 = 1.960; we do not reject the hypothesis (the sickness

rate is higher in workers than it is in farmers)

114
E x e r c i s e 5.42

When investigating the dependence of quantity y on quantity x the following data has
been measured:

xi 3.4 4.3 5.4 6.7 8.7 10.6

yi 4.5 5.8 6.8 8.1 10.5 12.7

Determine the regression function y = E1 + E2 x, calculate the point estimator for

y(5.4) and find 0.95 interval estimates for E1 , E2 , y(5.4).

R e s u l t: y = 0.77 + 1.12x; y(5.4) # 6.82; E1  <0.31;1.24>; E2  <1.05;1.19>;

y(5.4)  <6.41;7.23>

E x e r c i s e 5.43

To determine the dependence of this year's demand y on last year's demand x for a
certain type of goods the following data have been collected from 6 businessmen
(pieces):

xi 20 60 70 100 150 260

yi 50 60 60 120 230 320

Calculate the minimum variance unbiased point estimate and find a 95% interval
estimate for the coefficients of the regression line and for the value of this year's
demand at 110 pieces of last year's demand. With a 5% level of significance test the
hypothesis that E1 = 0 and determine the correlation coefficient.

R e s u l t: y =0.687 + 1.266x; E1  <-57.194; 58.568>; E2 <0.836; 1.696>;

y(110) # 140; y(110)  <106.55; 173.45>, resp. <51.50; 228.50>;

we do not reject the hypothesis; r = 0.97198

E x e r c i s e 5.44

The values y* (in thousands of items) of the demand for a certain type of goods at
prices x* (in thousands of CZK) are shown in the following table:

115
 xI* 100 110 140 160 200

yI* 120 89 56 41 22

Fit the data using a power regression function y* = Jx*G and find point and interval
estimates (using a 0.95 confidence) for the regression coefficients and for the
demand at price 120 CZK.

(Hint: take a logarithm of the power function.)

R e s u l t: ln y* = 15.64395 - 2.36035 ln x* ; y* = 6.224.106 x*-2.36;

ln J = E1  <13.95342; 17.33448>; G = E2  <-2.37817; -2.34253>;

y*(120) # 77; y*(120)  <69.8; 84.9> or <62.0; 95.6>

E x e r c i s e 5.45

The values y* (in thousands of CZK) of net sales at a company over the first six years
x* of operation are shown below:

xI* 1 2 3 4 5 6

yI* 112 149 238 354 580 867

Approximate the data using an exponential regression function y* = Jexp(Gx*) and

calculate point and interval estimates (using a 95 % confidence) of the regression
coefficients and make a forecast of the company's net sales in the seventh year of
operation.

(Hint: take a logarithm of the exponential function.)

R e s u l t: ln y* = 4.22798 + 0.4202x*; y* = 68.579exp(0.4202x*);

J = exp( E1 )  <59.41163; 79.15991>; G = E2  <0.38336; 0.45704>;

y*(7) # 1299.04; y*(7)  <1052.39; 1603.49>

Questions

1. Define a random sample and sample data.

2. What is meant by a sample characteristic and what are its properties?

116
3. Define the notion of a parameter and its types.

4. Define a point estimator and show point estimators for the basic number
characteristics.

5. Describe an interval estimator and a confidence interval for parameters.

6. Hoe does a change in the confidence level and the sample size affect the length
of a confidence interval?

7. Define a statistical hypothesis and describe its types.

8. What is a testing criterion and a critical range?

9. Describe type 1 and type 2 errors made at testing statistical hypotheses.

10. What is understood by regression analysis and a linear regression model?

11. What estimations and tests of statistical hypotheses are used in regression
analysis?

12. How is the suitability of a regression function appraised?

117
 BIBLIOGRAPHY

1. BOWERMAN, B. L., and O´CONNELL, R. T.: Applied Statistics; Improving

Business Processes. IRWIN, Chicago, 1997.

2. WONNACOT, T.H., and WONNACOT, R. J.: Introductory Statistic for Business

and Economics. John Wiley & Sons, Inc., New York, 1993.

3. SPRINTHALL, R. C.: Basic Statistical Analysis (5th ed.). Allyn and Bacon, Boston,
1997.

4. ACZEL, A. D.: Complete Business Statistics. IRWIN, Chicago, 1989.

5. TRIOLA, M. F.: Elementary Statistics. B/C Publishing Comp., Redwood City,

1989.

6. AMEMIYA, T.: Introduction to Statistics & Econometrics. Harvard University

Press, 1994.

7. BERENSON, M.: Business Statistics; A First Course. New York, Prentice Hall
1997.

8. WEIERS, R. M.: Introduction to Business Statistics. 3. e. Brooks/Cole Publishing

1997.

9. SEGER, J., and HINDLS, R.: Statistické metody v tržním hospodáĜství. Victoria
Publishing, Praha, 1995.

10. ŠIKULOVÁ, M., and KARPÍŠEK, Z.: Matematika IV (PravdČpodobnost a

matematická statistika). ES VUT, Brno, 1996.

11. LIKEŠ, J., CYHELSKÝ, L., and HINDLS, R.: Úvod do statistiky a pravdČpodob-
nosti (Statistika A). VŠE, Praha, 1995.

12. JAROŠOVÁ, E.: Statistika B. ěešené pĜíklady. VŠE, Praha, 1994.

13. MELOUN, M., and MILITKÝ, J.: Statistické zpracování experimentálních

dat. PLUS, Praha, 1994.

14. SEBEROVÁ, H.: Statistika I, II. VVŠ PV, Vyškov, 1995.

15. CIPRA, T.: Analýza þasových Ĝad s aplikacemi v ekonomii. SNTL/Alfa, Praha,
1986.

118
 STATISTICAL TABLES

T1 Distribution function )(u) of the standard normal distribution N(0;1)

u 0 1 2 3 4 5 6 7 8 9
0.0 0.50000 50399 50798 51197 51596 51994 52392 52791 53188 53586
0.1 53983 54380 54776 55172 55567 55962 56356 56750 57143 57535
0.2 57926 58317 58707 59096 59484 59871 60257 60642 61026 61409
0.3 61791 62172 62552 62930 63307 63683 64058 64431 64803 65173
0.4 65542 65910 66276 66640 67003 67365 67724 68082 68439 68793
0.5 69146 69498 69847 70195 70540 70884 71226 71566 71904 72241
0.6 72575 72907 73237 73565 73892 74216 74537 74857 75175 75490
0.7 75804 76115 76424 76731 77035 77337 77637 77935 78231 78524
0.8 78815 79103 79389 79673 79955 80234 80511 80785 81057 81327
0.9 81594 81859 82121 82382 82639 82894 83147 83398 83646 83891
1.0 84135 84375 84614 84850 85083 85314 85543 85769 85993 86214
1.1 86433 86650 86864 87076 87286 87493 87698 87900 88100 88298
1.2 88493 88686 88877 89065 89251 89435 89617 89796 89973 90147
1.3 90320 90490 90658 90824 90988 91149 91309 91466 91621 91774
1.4 91924 92073 92220 92364 92507 92647 92786 92922 93056 93189
1.5 93319 93448 93574 93699 93822 93943 94062 94179 94295 94408
1.6 94520 94630 94738 94845 94950 95053 95154 95254 95352 95449
1.7 95543 95637 95728 95819 95907 95994 96080 96164 96246 96327
1.8 96407 96485 96562 96638 96712 96784 96856 96926 96995 97062
1.9 97128 97193 97257 97320 97381 97441 97500 97558 97615 97670
2.0 97725 97778 97831 97882 97932 97982 98030 98077 98124 98169
2.1 98214 98257 98300 98341 98382 98422 98461 98500 98537 98574
2.2 98610 98645 98679 98713 98745 98778 98809 98840 98870 98899
2.3 98928 98956 98983 99010 99036 99061 99086 99111 99134 99158
2.4 99180 99202 99224 99245 99266 99286 99305 99324 99343 99361
2.5 99379 99396 99413 99430 99446 99461 99477 99492 99506 99520
2.6 99534 99547 99560 99573 99585 99598 99609 99621 99632 99643
2.7 99653 99664 99674 99683 99693 99702 99711 99720 99728 99736
2.8 99744 99752 99760 99767 99774 99781 99788 99795 99801 99807
2.9 99813 99819 99825 99831 99836 99841 99846 99851 99856 99861
3.0 99865 99869 99874 99878 99882 99886 99889 99893 99896 99900
3.1 99903 99906 99910 99913 99916 99918 99921 99924 99926 99929
3.2 99931 99934 99936 99938 99940 99942 99944 99946 99948 99950
3.3 99952 99953 99955 99957 99958 99960 99961 99962 99964 99965
3.4 99966 99968 99969 99970 99971 99972 99973 99974 99975 99976
3.5 99977 99978 99978 99979 99980 99981 99981 99982 99983 99983
3.6 99984 99985 99985 99986 99986 99987 99987 99988 99988 99989
3.7 99989 99990 99990 99990 99991 99991 99992 99992 99992 99992
3.8 99993 99993 99993 99994 99994 99994 99994 99995 99995 99995
3.9 99995 99995 99996 99996 99996 99996 99996 99996 99997 99997
4.00 99997 4.10 99998 4.20 99999 4.30 99999 4.40 99999 4.50 99999

119
 T2 Quantiles tP of the Student distribution S(k)

P 0.95 0.975 0.99 0.995 0.999 0.9995

k
1 6.314 12.706 31.821 63.656 318.289 636.578
2 2.920 4.303 6.965 9.925 22.328 31.600
3 2.353 3.182 4.541 5.841 10.214 12.924
4 2.132 2.776 3.747 4.604 7.173 8.610
5 2.015 2.571 3.365 4.032 5.894 6.869
6 1.943 2.447 3.143 3.707 5.208 5.959
7 1.895 2.365 2.998 3.499 4.785 5.408
8 1.860 2.306 2.896 3.355 4.501 5.041
9 1.833 2.262 2.821 3.250 4.297 4.781
10 1.812 2.228 2.764 3.169 4.144 4.587
11 1.796 2.201 2.718 3.106 4.025 4.437
12 1.782 2.179 2.681 3.055 3.930 4.318
13 1.771 2.160 2.650 3.012 3.852 4.221
14 1.761 2.145 2.624 2.977 3.787 4.140
15 1.753 2.131 2.602 2.947 3.733 4.073
16 1.746 2.120 2.583 2.921 3.686 4.015
17 1.740 2.110 2.567 2.898 3.646 3.965
18 1.734 2.101 2.552 2.878 3.610 3.922
19 1.729 2.093 2.539 2.861 3.579 3.883
20 1.725 2.086 2.528 2.845 3.552 3.850
21 1.721 2.080 2.518 2.831 3.527 3.819
22 1.717 2.074 2.508 2.819 3.505 3.792
23 1.714 2.069 2.500 2.807 3.485 3.768
24 1.711 2.064 2.492 2.797 3.467 3.745
25 1.708 2.060 2.485 2.787 3.450 3.725
26 1.706 2.056 2.479 2.779 3.435 3.707
27 1.703 2.052 2.473 2.771 3.421 3.689
28 1.701 2.048 2.467 2.763 3.408 3.674
29 1.699 2.045 2.462 2.756 3.396 3.660
30 1.697 2.042 2.457 2.750 3.385 3.646
35 1.690 2.030 2.438 2.724 3.340 3.591
40 1.684 2.021 2.423 2.704 3.307 3.551
45 1.679 2.014 2.412 2.690 3.281 3.520
50 1.676 2.009 2.403 2.678 3.261 3.496
60 1.671 2.000 2.390 2.660 3.232 3.460
70 1.667 1.994 2.381 2.648 3.211 3.435
80 1.664 1.990 2.374 2.639 3.195 3.416
90 1.662 1.987 2.368 2.632 3.183 3.402
100 1.660 1.984 2.364 2.626 3.174 3.390
120 1.658 1.980 2.358 2.617 3.160 3.373
140 1.656 1.977 2.353 2.611 3.149 3.361
160 1.654 1.975 2.350 2.607 3.142 3.352
180 1.653 1.973 2.347 2.603 3.136 3.345
200 1.653 1.972 2.345 2.601 3.131 3.340
300 1.650 1.968 2.339 2.592 3.118 3.323
400 1.649 1.966 2.336 2.588 3.111 3.315
500 1.648 1.965 2.334 2.586 3.107 3.310
1000 1.646 1.962 2.330 2.581 3.098 3.300
f 1.645 1.960 2.326 2.576 3.090 3.290

120
 2
T3 Quantiles FP2 of the Pearson distribution F (k)

P 0.005 0.01 0.025 0.05 0.95 0.975 0.99 0.995

k
1 0.000 0.000 0.001 0.004 3.841 5.024 6.635 7.879
2 0.010 0.020 0.051 0.103 5.991 7.378 9.210 10.597
3 0.072 0.115 0.216 0.352 7.815 9.348 11.345 12.838
4 0.207 0.297 0.484 0.711 9.488 11.143 13.277 14.860
5 0.412 0.554 0.831 1.145 11.070 12.832 15.086 16.750
6 0.676 0.872 1.237 1.635 12.592 14.449 16.812 18.548
7 0.989 1.239 1.690 2.167 14.067 16.013 18.475 20.278
8 1.344 1.647 2.180 2.733 15.507 17.535 20.090 21.955
9 1.735 2.088 2.700 3.325 16.919 19.023 21.666 23.589
10 2.156 2.558 3.247 3.940 18.307 20.483 23.209 25.188
11 2.603 3.053 3.816 4.575 19.675 21.920 24.725 26.757
12 3.074 3.571 4.404 5.226 21.026 23.337 26.217 28.300
13 3.565 4.107 5.009 5.892 22.362 24.736 27.688 29.819
14 4.075 4.660 5.629 6.571 23.685 26.119 29.141 31.319
15 4.601 5.229 6.262 7.261 24.996 27.488 30.578 32.801
16 5.142 5.812 6.908 7.962 26.296 28.845 32.000 34.267
17 5.697 6.408 7.564 8.672 27.587 30.191 33.409 35.718
18 6.265 7.015 8.231 9.390 28.869 31.526 34.805 37.156
19 6.844 7.633 8.907 10.117 30.144 32.852 36.191 38.582
20 7.434 8.260 9.591 10.851 31.410 34.170 37.566 39.997
21 8.034 8.897 10.283 11.591 32.671 35.479 38.932 41.401
22 8.643 9.542 10.982 12.338 33.924 36.781 40.289 42.796
23 9.260 10.196 11.689 13.091 35.172 38.076 41.638 44.181
24 9.886 10.856 12.401 13.848 36.415 39.364 42.980 45.558
25 10.520 11.524 13.120 14.611 37.652 40.646 44.314 46.928
26 11.160 12.198 13.844 15.379 38.885 41.923 45.642 48.290
27 11.808 12.878 14.573 16.151 40.113 43.195 46.963 49.645
28 12.461 13.565 15.308 16.928 41.337 44.461 48.278 50.994
29 13.121 14.256 16.047 17.708 42.557 45.722 49.588 52.335
30 13.787 14.953 16.791 18.493 43.773 46.979 50.892 53.672
31 14.458 15.655 17.539 19.281 44.985 48.232 52.191 55.002
32 15.134 16.362 18.291 20.072 46.194 49.480 53.486 56.328
33 15.815 17.073 19.047 20.867 47.400 50.725 54.775 57.648
34 16.501 17.789 19.806 21.664 48.602 51.966 56.061 58.964
35 17.192 18.509 20.569 22.465 49.802 53.203 57.342 60.275
36 17.887 19.233 21.336 23.269 50.998 54.437 58.619 61.581
37 18.586 19.960 22.106 24.075 52.192 55.668 59.893 62.883
38 19.289 20.691 22.878 24.884 53.384 56.895 61.162 64.181
39 19.996 21.426 23.654 25.695 54.572 58.120 62.428 65.475
40 20.707 22.164 24.433 26.509 55.758 59.342 63.691 66.766
41 21.421 22.906 25.215 27.326 56.942 60.561 64.950 68.053
42 22.138 23.650 25.999 28.144 58.124 61.777 66.206 69.336
43 22.860 24.398 26.785 28.965 59.304 62.990 67.459 70.616
44 23.584 25.148 27.575 29.787 60.481 64.201 68.710 71.892
45 24.311 25.901 28.366 30.612 61.656 65.410 69.957 73.166

121
 2
T3 Quantiles FP2 of the Pearson distribution F (k) (continued)

P 0.005 0.01 0.025 0.05 0.95 0.975 0.99 0.995

k
46 25.041 26.657 29.160 31.439 62.830 66.616 71.201 74.437
47 25.775 27.416 29.956 32.268 64.001 67.821 72.443 75.704
48 26.511 28.177 30.754 33.098 65.171 69.023 73.683 76.969
49 27.249 28.941 31.555 33.930 66.339 70.222 74.919 78.231
50 27.991 29.707 32.357 34.764 67.505 71.420 76.154 79.490
51 28.735 30.475 33.162 35.600 68.669 72.616 77.386 80.746
52 29.481 31.246 33.968 36.437 69.832 73.810 78.616 82.001
53 30.230 32.019 34.776 37.276 70.993 75.002 79.843 83.253
54 30.981 32.793 35.586 38.116 72.153 76.192 81.069 84.502
55 31.735 33.571 36.398 38.958 73.311 77.380 82.292 85.749
56 32.491 34.350 37.212 39.801 74.468 78.567 83.514 86.994
57 33.248 35.131 38.027 40.646 75.624 79.752 84.733 88.237
58 34.008 35.914 38.844 41.492 76.778 80.936 85.950 89.477
59 34.770 36.698 39.662 42.339 77.930 82.117 87.166 90.715
60 35.534 37.485 40.482 43.188 79.082 83.298 88.379 91.952
61 36.300 38.273 41.303 44.038 80.232 84.476 89.591 93.186
62 37.068 39.063 42.126 44.889 81.381 85.654 90.802 94.419
63 37.838 39.855 42.950 45.741 82.529 86.830 92.010 95.649
64 38.610 40.649 43.776 46.595 83.675 88.004 93.217 96.878
65 39.383 41.444 44.603 47.450 84.821 89.177 94.422 98.105
66 40.158 42.240 45.431 48.305 85.965 90.349 95.626 99.330
67 40.935 43.038 46.261 49.162 87.108 91.519 96.828 100.554
68 41.714 43.838 47.092 50.020 88.250 92.688 98.028 101.776
69 42.493 44.639 47.924 50.879 89.391 93.856 99.227 102.996
70 43.275 45.442 48.758 51.739 90.531 95.023 100.425 104.215
71 44.058 46.246 49.592 52.600 91.670 96.189 101.621 105.432
72 44.843 47.051 50.428 53.462 92.808 97.353 102.816 106.647
73 45.629 47.858 51.265 54.325 93.945 98.516 104.010 107.862
74 46.417 48.666 52.103 55.189 95.081 99.678 105.202 109.074
75 47.206 49.475 52.942 56.054 96.217 100.839 106.393 110.285
80 51.172 53.540 57.153 60.391 101.879 106.629 112.329 116.321
85 55.170 57.634 61.389 64.749 107.522 112.393 118.236 122.324
90 59.196 61.754 65.647 69.126 113.145 118.136 124.116 128.299
95 63.250 65.898 69.925 73.520 118.752 123.858 129.973 134.247
100 67.328 70.065 74.222 77.929 124.342 129.561 135.807 140.170
110 75.550 78.458 82.867 86.792 135.480 140.916 147.414 151.948
120 83.852 86.923 91.573 95.705 146.567 152.211 158.950 163.648
130 92.223 95.451 100.331 104.662 157.610 163.453 170.423 175.278
150 109.142 112.668 117.985 122.692 179.581 185.800 193.207 198.360
200 152.241 156.432 162.728 168.279 233.994 241.058 249.445 255.264
500 422.303 429.387 439.936 449.147 553.127 563.851 576.493 585.206
1000 888.563 898.912 914.257 927.594 1074.68 1089.53 1106.97 1118.95

122
 T4 Quantiles FP of the Fisher – Snedecor distribution F(k1,k2) for P = 0.95
k1
1 2 3 4 5 6 8 12 24 f
k2
1 161.446 199.499 215.707 224.583 230.160 233.988 238.884 243.905 249.052 254.313
2 18.513 19.000 19.164 19.247 19.296 19.329 19.371 19.412 19.454 19.496
3 10.128 9.552 9.277 9.117 9.013 8.941 8.845 8.745 8.638 8.526
4 7.709 6.944 6.591 6.388 6.256 6.163 6.041 5.912 5.774 5.628
5 6.608 5.786 5.409 5.192 5.050 4.950 4.818 4.678 4.527 4.365
6 5.987 5.143 4.757 4.534 4.387 4.284 4.147 4.000 3.841 3.669
7 5.591 4.737 4.347 4.120 3.972 3.866 3.726 3.575 3.410 3.230
8 5.318 4.459 4.066 3.838 3.688 3.581 3.438 3.284 3.115 2.928
9 5.117 4.256 3.863 3.633 3.482 3.374 3.230 3.073 2.900 2.707
10 4.965 4.103 3.708 3.478 3.326 3.217 3.072 2.913 2.737 2.538
11 4.844 3.982 3.587 3.357 3.204 3.095 2.948 2.788 2.609 2.404
12 4.747 3.885 3.490 3.259 3.106 2.996 2.849 2.687 2.505 2.296
13 4.667 3.806 3.411 3.179 3.025 2.915 2.767 2.604 2.420 2.206
14 4.600 3.739 3.344 3.112 2.958 2.848 2.699 2.534 2.349 2.131
15 4.543 3.682 3.287 3.056 2.901 2.790 2.641 2.475 2.288 2.066
16 4.494 3.634 3.239 3.007 2.852 2.741 2.591 2.425 2.235 2.010
17 4.451 3.592 3.197 2.965 2.810 2.699 2.548 2.381 2.190 1.960
18 4.414 3.555 3.160 2.928 2.773 2.661 2.510 2.342 2.150 1.917
19 4.381 3.522 3.127 2.895 2.740 2.628 2.477 2.308 2.114 1.878
20 4.351 3.493 3.098 2.866 2.711 2.599 2.447 2.278 2.082 1.843
21 4.325 3.467 3.072 2.840 2.685 2.573 2.420 2.250 2.054 1.812
22 4.301 3.443 3.049 2.817 2.661 2.549 2.397 2.226 2.028 1.783
23 4.279 3.422 3.028 2.796 2.640 2.528 2.375 2.204 2.005 1.757
24 4.260 3.403 3.009 2.776 2.621 2.508 2.355 2.183 1.984 1.733
25 4.242 3.385 2.991 2.759 2.603 2.490 2.337 2.165 1.964 1.711
26 4.225 3.369 2.975 2.743 2.587 2.474 2.321 2.148 1.946 1.691
27 4.210 3.354 2.960 2.728 2.572 2.459 2.305 2.132 1.930 1.672
28 4.196 3.340 2.947 2.714 2.558 2.445 2.291 2.118 1.915 1.654
29 4.183 3.328 2.934 2.701 2.545 2.432 2.278 2.104 1.901 1.638
30 4.171 3.316 2.922 2.690 2.534 2.421 2.266 2.092 1.887 1.622
35 4.121 3.267 2.874 2.641 2.485 2.372 2.217 2.041 1.833 1.558
40 4.085 3.232 2.839 2.606 2.449 2.336 2.180 2.003 1.793 1.509
45 4.057 3.204 2.812 2.579 2.422 2.308 2.152 1.974 1.762 1.470
50 4.034 3.183 2.790 2.557 2.400 2.286 2.130 1.952 1.737 1.438
55 4.016 3.165 2.773 2.540 2.383 2.269 2.112 1.933 1.717 1.412
60 4.001 3.150 2.758 2.525 2.368 2.254 2.097 1.917 1.700 1.389
70 3.978 3.128 2.736 2.503 2.346 2.231 2.074 1.893 1.674 1.353
80 3.960 3.111 2.719 2.486 2.329 2.214 2.056 1.875 1.654 1.325
90 3.947 3.098 2.706 2.473 2.316 2.201 2.043 1.861 1.639 1.302
100 3.936 3.087 2.696 2.463 2.305 2.191 2.032 1.850 1.627 1.283
120 3.920 3.072 2.680 2.447 2.290 2.175 2.016 1.834 1.608 1.254
150 3.904 3.056 2.665 2.432 2.274 2.160 2.001 1.817 1.590 1.223
250 3.879 3.032 2.641 2.408 2.250 2.135 1.976 1.791 1.561 1.166
500 3.860 3.014 2.623 2.390 2.232 2.117 1.957 1.772 1.539 1.113
f 3.841 2.996 2.605 2.372 2.214 2.099 1.938 1.752 1.517 1.000

123
 T4 Quantiles FP of the Fisher – Snedecor distribution F(k1,k2) for P = 0.99
k1
1 2 3 4 5 6 8 12 24 f
k2
1 4052.18 4999.34 5403.53 5624.26 5763.96 5858.95 5980.95 6106.68 6234.27 6365.59
2 98.502 99.000 99.164 99.251 99.302 99.331 99.375 99.419 99.455 99.499
3 34.116 30.816 29.457 28.710 28.237 27.911 27.489 27.052 26.597 26.125
4 21.198 18.000 16.694 15.977 15.522 15.207 14.799 14.374 13.929 13.463
5 16.258 13.274 12.060 11.392 10.967 10.672 10.289 9.888 9.466 9.020
6 13.745 10.925 9.780 9.148 8.746 8.466 8.102 7.718 7.313 6.880
7 12.246 9.547 8.451 7.847 7.460 7.191 6.840 6.469 6.074 5.650
8 11.259 8.649 7.591 7.006 6.632 6.371 6.029 5.667 5.279 4.859
9 10.562 8.022 6.992 6.422 6.057 5.802 5.467 5.111 4.729 4.311
10 10.044 7.559 6.552 5.994 5.636 5.386 5.057 4.706 4.327 3.909
11 9.646 7.206 6.217 5.668 5.316 5.069 4.744 4.397 4.021 3.602
12 9.330 6.927 5.953 5.412 5.064 4.821 4.499 4.155 3.780 3.361
13 9.074 6.701 5.739 5.205 4.862 4.620 4.302 3.960 3.587 3.165
14 8.862 6.515 5.564 5.035 4.695 4.456 4.140 3.800 3.427 3.004
15 8.683 6.359 5.417 4.893 4.556 4.318 4.004 3.666 3.294 2.868
16 8.531 6.226 5.292 4.773 4.437 4.202 3.890 3.553 3.181 2.753
17 8.400 6.112 5.185 4.669 4.336 4.101 3.791 3.455 3.083 2.653
18 8.285 6.013 5.092 4.579 4.248 4.015 3.705 3.371 2.999 2.566
19 8.185 5.926 5.010 4.500 4.171 3.939 3.631 3.297 2.925 2.489
20 8.096 5.849 4.938 4.431 4.103 3.871 3.564 3.231 2.859 2.421
21 8.017 5.780 4.874 4.369 4.042 3.812 3.506 3.173 2.801 2.360
22 7.945 5.719 4.817 4.313 3.988 3.758 3.453 3.121 2.749 2.305
23 7.881 5.664 4.765 4.264 3.939 3.710 3.406 3.074 2.702 2.256
24 7.823 5.614 4.718 4.218 3.895 3.667 3.363 3.032 2.659 2.211
25 7.770 5.568 4.675 4.177 3.855 3.627 3.324 2.993 2.620 2.169
26 7.721 5.526 4.637 4.140 3.818 3.591 3.288 2.958 2.585 2.131
27 7.677 5.488 4.601 4.106 3.785 3.558 3.256 2.926 2.552 2.097
28 7.636 5.453 4.568 4.074 3.754 3.528 3.226 2.896 2.522 2.064
29 7.598 5.420 4.538 4.045 3.725 3.499 3.198 2.868 2.495 2.034
30 7.562 5.390 4.510 4.018 3.699 3.473 3.173 2.843 2.469 2.006
35 7.419 5.268 4.396 3.908 3.592 3.368 3.069 2.740 2.364 1.891
40 7.314 5.178 4.313 3.828 3.514 3.291 2.993 2.665 2.288 1.805
45 7.234 5.110 4.249 3.767 3.454 3.232 2.935 2.608 2.230 1.737
50 7.171 5.057 4.199 3.720 3.408 3.186 2.890 2.563 2.183 1.683
55 7.119 5.013 4.159 3.681 3.370 3.149 2.853 2.526 2.146 1.638
60 7.077 4.977 4.126 3.649 3.339 3.119 2.823 2.496 2.115 1.601
70 7.011 4.922 4.074 3.600 3.291 3.071 2.777 2.450 2.067 1.540
80 6.963 4.881 4.036 3.563 3.255 3.036 2.742 2.415 2.032 1.494
90 6.925 4.849 4.007 3.535 3.228 3.009 2.715 2.389 2.004 1.457
100 6.895 4.824 3.984 3.513 3.206 2.988 2.694 2.368 1.983 1.427
120 6.851 4.787 3.949 3.480 3.174 2.956 2.663 2.336 1.950 1.381
150 6.807 4.749 3.915 3.447 3.142 2.924 2.632 2.305 1.918 1.331
250 6.737 4.691 3.861 3.395 3.091 2.875 2.583 2.256 1.867 1.244
500 6.686 4.648 3.821 3.357 3.054 2.838 2.547 2.220 1.829 1.164
f 6.635 4.605 3.782 3.319 3.017 2.802 2.511 2.185 1.791 1.000

124