O P E R A T I O N S R E S E A R C H A N D D E C I S I O N S

No. 1 2011

Roman SZOSTEK*
AN EFFECTIVE SYSTEM
OF SPORTS COMPETITION MANAGEMENT
An innovatory system of managing sports competitions has been presented. Its advantages with
regard to other currently used systems are discussed. A theorem connected with such a system has
been presented in the last section of the paper. Sports competitions aim to establish a ranking of the
participating teams. This consists of sorting teams according to a quality which can be thought of as
the ability to win matches. Direct measurement of this quality is not possible, since the ability to win
matches depends on a great variety of factors being difficult to determine. Nevertheless, it is possible
to compare any two teams if they play a match. These matches are played under normal rules. In turn,
all the rules valid during sports competitions, outside the matches, make a system of sport competi-
tion. Sorting sports teams differs from typical problems of sorting. The result of a comparison of
teams is sometimes misleading. It happens that a team with a greater ability to win matches loses
a match to a team with a smaller ability to win matches. Thus, the problem of sorting teams is a prob-
abilistic problem. Due to this reason, traditional sorting methods are ineffective in terms of managing
sports competitions.
Keywords: sports games, championships, system of sports competitions, probabilistic sorting, computa-
tional complexity
1. Introduction
The aim of sports competitions is to rank teams or competitors taking part in
sports games. Thus, the problem is how to sort teams by their ability to win matches.
However, it is not possible to directly measure this characteristic because such ability
depends on many factors that are difficult to define. It is possible, however, to com-
pare two randomly chosen teams provided they have played a match. All the rules
__________________________
*The Faculty of Management Department of Quantitative Methods in Economics, Rzeszw Univer-
sity of Technology, Wincentego Pola 2, 35-959 Rzeszw, e-mail: [email protected]
R. SZOSTEK

66
applied in various competitions are systems of managing sports competitions. Thus
a system of sports competitions consists of the rules which decide who plays against
whom, in what sequence and where, as well as the criteria that decide the teams places
in the final ranking list.
Sorting sports teams differs from standard sorting problems, due to the fact that
the result of comparing two teams is sometimes incorrect. It does happen that a team
with a greater ability to win gets defeated by a team whose winning potential is un-
doubtedly lower. Thus the problem of sorting teams is a probabilistic one. This is
a reason why traditional methods of sorting do not work in the case of the manage-
ment of sports competitions.
2. Formulation of the problem
There are n objects d
1
, d
2
, ..., d
n
D. Some numerical value called quality is as-
signed to each object. The object d
i
has quality x
i
R (i = 1, 2, ..., n). It is not possible
to measure the quality of any of the objects. However, comparison of the quality of
any two objects is possible (e.g. by the result of a game). As a result of comparing two
objects d
i
and d
j
, the relation established is d
i
d
j
(d
j
wins against d
i
) or d
j
d
i

(d
i
wins against d
j
) such that

,
Pr ( ) Pr ( )
i j i j ij j i ij
i j
x x d d p d d q = =

(1)
where: p
ij
1/2 is the probability that the highest quality team wins, q
ij
1/2 is the
probability that the lowest quality team wins, p
ij
+ q
ij
= 1.
We assume that the probability p
ij
is a non-decreasing function of the distance be-
tween the qualities x
i
and x
j
. Consequently, we have
(| |)
ij ji i j
p p f x x = = (2)
0 ( ) ( ) a b f a f b (3)

1
(0)
2
f = (4)
Objects belonging to the set D are to be sorted with respect to quality according to
the conditions stated above.
Management of the teaching process

67
3. Discussion of the problem
A management system for sports competitions is a solution to the problem of
probabilistic sorting. It consists of sorting objects under conditions where the only tool
that we possess is probabilistic pan scales. Their properties are presented in conditions
(1)(4). The probabilistic character of the scales results in the fact that the relation
established between the same pair of objects can differ according to different weight-
ings.
Condition (1) guarantees that if two objects have different qualities, the probabil-
ity p
ij
that the relation between them will be consistent with the relation between the
qualities is greater than the probability q
ij
that the established relation will not be con-
sistent with the relation between the qualities. This dependence contains also an as-
sumption that the result of the comparison of two objects never indicates an equality
relation between them (p
ij
+ q
ij
= 1).
Condition (2) guarantees that the probability p
ij
that the relation between two
compared objects will be consistent with the relation between their qualities depends
only on the extent of the difference between these qualities. Therefore, the order of the
compared objects is not important, thus p
ij
= p
ji
.
Condition (3) guarantees that the probability p
ij
that the relation between two
compared objects will be consistent with the relation between their qualities is increas-
ing in the difference between these qualities, i.e. the result of weighing is more often
true when two objects differ greatly than when two objects are more similar.
Condition (4) guarantees that the result of the comparison of two objects of the
same quality does not favour either of the objects.

Fig. 1. An example of the function p
ij
An example function p
ij
is presented in Fig. 1. It is obvious that under probabilistic
sorting no method guarantees obtaining exactly the same ordering as the ordering of
the qualities of the sorted objects. Appropriate methods may merely guarantee a high
probability of obtaining such an order.
R. SZOSTEK

68
4. Solutions applied in sport
Two common systems of managing sports games are presently used all over the
world. The former is the round-robin system (the peer-to-peer system) known as the
English system. The other one is the knockout system. Also, various variants combin-
ing these two systems are applied.
In the round-robin system all possible pairs of participating teams are compared.
Each pair of teams is compared a number of times set in advance. One of the disad-
vantages of the system is the need for making numerous comparisons. The number of
comparisons is a multiple of n(n 1)/2, where n is the number of teams. Thus, the
computational complexity of this method is polynomial. Another disadvantage of this
method is the fact that one cannot freely arrange the number of all planned compari-
sons during the process of sorting. This method is also ineffective because teams that
differ considerably are compared unnecessarily often. Another drawback concerns the
impossibility of a precise, that is to say, frequent comparison of teams that are similar.
It is the result of the comparison of such teams that is often incorrect and needs check-
ing several times.
The knockout system appears in several variants. In the basic variant, a team that
has lost a match is eliminated from games. Thus, after each sequence of matches, only
half of the teams remain. The team that does not lose any match becomes the winner.
The number of comparisons made equals n1. Various combinations of the round-
robin system and the knockout system are used. One disadvantage of a simple knock-
out system is the fact that it does not allow arranging all the teams yet merely deter-
mines a winner. Besides, this system assumes that the probability p
ij
is constant and
equals 1.
In chess competitions, the Swiss system is used. The drawback of this system for
ranking chess players concerns the fact that it does not guarantee the possibility of
more precise, that is, more frequent comparisons of players that have similar qualities,
since two players can be compared only once.
5. Algorithm for probabilistic sorting
The proposed algorithm of probabilistic sorting has the following properties.
1. Sorting is done in stages. At each stage objects are placed in a sequence.
Neighbouring objects are compared, i.e. the first one with the second one, the third
one with the fourth one. The first arrangement of objects, the so-called starting ar-
rangement, may be made at random.
Management of the teaching process

69
2. After each comparison, a new arrangement of objects in the sequence is made.
It is established in such a way that if d
i
d
j
, then the object d
j
stays in its place or
moves up the sequence, whereas the object d
i
stays in its place or moves down the
sequence.
3. Every object has its own level. Thanks to this fact, objects can obtain a differing
number of points according to the results of a comparison at this level. At the begin-
ning, all the objects are placed on the same level, which is the lowest one.
4. After each comparison, objects move to new levels. They move in such a way
that if d
i
d
j
, then object d
j
moves to a level not lower than its current one, whereas
the object d
i
moves to a level not higher than its current one (according to this proce-
dure, objects will change levels at various stages of the algorithm).
5. After each comparison, objects gain points. If d
i
d
j
, then d
j
gains more points
than d
i
. Additionally, more points are gained by objects that are at higher levels than
those at lower levels.
Many variants of this algorithm can be constructed in accordance with the above
rules. If the number of objects is odd, the sequence should be completed with a virtual
object of a quality that is smaller than any of the qualities of the teams.
A formal definition of the algorithm for an even number of objects is given below.
The present sequence of objects is put into pairs which are numbered 1, 2, , n/2.
Let
k
i
a denote the number of the pair containing object d
i
in the k-th sequence.
{1, 2, 3, ..., /2}
k
i
a n is the number of the pair in the sequence containing object d
i
,
{1, 2, 3, ..., } k K is the number of the sequence,
{1, 2, 3, ..., } i n is the number of the object.
For each k,
1 2
( , , ..., )
k k k
n
a a a is a permutation of the elements of the set {1, 1, 2, 2,
3, 3, , n/2, n/2}.. If in the k-th series of comparisons, the two objects d
i
and d
j
are
compared (i.e.
k k
i j
a a = ) and it was established that d
i
d
j
, then the position of these
objects in the next sequence satisfies the condition
1 1
1 /2.
k k k k
j j i i
a a a a n
+ +
=
One of M levels is associated with each pair of objects. Let
k
i
b denote the level of
the pair containing object d
i
in the k-th sequence:
{1, 2, 3, ..., }
k
i
b M is the level of the pair containing object d
i
in the k-th sequence.
All the pairs in the first sequence have the lowest level (level 1). In all the follow-
ing sequences the levels are assigned to the pairs in such a way that if, in the k-th se-
quence, it turns out that d
i
d
j
for two comparable objects, then
1 1
1 .
k k k k
i i j j
b b b b M
+ +
=
R. SZOSTEK

70
Beginning from a random sequence, after each set of comparisons of pairs of ob-
jects, the objects obtain points. The number of points obtained depends on the pairs
level and the result of the comparison conducted in the way described above.

Fig. 2. The sorting algorithm the first round

Fig. 3. The sorting algorithm the second round

Fig. 4. The sorting algorithm the third round

Fig. 5. The sorting algorithm from the fourth round onwards
Figures 25 illustrate an example algorithm for 16 objects. In each sequence, the
objects form eight pairs a basis of arranging matches. Each pair is marked by a cir-
cle. The number of a circle stands for its place in the sequence and not for a particular
pair of objects.
First, the starting arrangement of the objects is made, as well as comparisons of
pairs in this first sequence. After eight comparisons based on this first sequence, eight
Management of the teaching process

71
objects that won and eight objects defeated are obtained. Then all the objects are ar-
ranged in the next sequence as shown in the figures. The upper curves indicate the
movement of the objects which won in the previous round, while the lower curves
indicate the movement of the objects which lost in the previous round. As can be seen
in the figures, the winning objects move up the sequence, whereas the objects that lost
move down the sequence. Thanks to this, objects of similar characteristics are more
likely to be compared in the following rounds. It was originally assumed that a draw is
out of the question. That is why, if a match ends in a draw, its result is decided by
extra time or randomly.
Each object is on one of the levels: A, B, C or D. The lowest level is A and the
highest one is D. At first, all the objects are on level A. This is shown in Fig. 2, which
represents the first sequence. After the comparisons based on the first sequence, the
objects which won move to the higher level B, while the objects which lost stay on
level A. New levels are shown in Fig. 3, which represents the second sequence. Now,
objects which won again move to a higher level, i.e. from A to B, or from B to C. The
objects which lost remain on their levels. The new levels are shown in Fig. 4, which
represents the third sequence. In the case of sequence three, the procedure is the same.
From the fourth sequence onwards, the procedure changes. Now only the level an
object occupies will change. This occurs in the way shown in Fig. 5. For instance, if
one looks at the objects in pair 1, they are on level D. After the objects are compared,
the object that won stays in pair 1, whereas the one that lost moves to pair 3. Thus,
a winning object stays on the same level or advances to a higher one. On the other
hand, an object that lost remains on the same level or falls down to a lower level. Fol-
lowing rounds are held in the same manner as the fourth round. A stop condition is
established at the beginning of the procedure.
After each round, objects get points for the results achieved. The number of points ob-
tained depends on the level at which a given object is placed. An example point system is
presented in Table 1. Objects placed on a higher level obtain more points than objects
from lower levels. To obtain the final ranking, the points obtained by each object in all the
rounds are added. The object that gets the most points is given first place.
Table 1. An example point scheme
The level of
an object
Points for an object
Win Lose
The highest D 4 3
C 3 2
B 2 1
The lowest A 1 0
In the conducted simulations, very good results were obtained when the objects
were ordered after the fourth round. In the first three rounds, the objects competed
R. SZOSTEK

72
simply to obtain a higher level. The number of comparisons required under the pro-
posed algorithm for probabilistic sorting equals n(t + log
2
n1)/2, where t is the number
of repetitions of the procedure shown in Fig. 5. It seems that it is sufficient that t is of
order log
2
n, thus the complexity of the algorithm is O(nlog
2
n), which is the same as
the mean complexity of the quicksort algorithm.
6. Theorem of neat systems
Let X and Y be pairs in a sequence. The notation X > Y will be adopted when the
pair X is located higher than the pair Y, i.e. following the notation used in the example
above the number of pair X is lower than that of pair Y.
6.1. Definition of a neat system
A neat system is a system where, for each X and Y, if X >Y, then in the next sequence
the winner from pair X cannot be in a lower pair than the winner from pair Y. Similarly,
the loser from pair X cannot be in a lower pair than the loser from pair Y.
The notation used is as follows:
S
1
set of systems where, in each round, winners (from the previous round) can
only compete with winners and losers (from the previous round) only with losers.
S
2
set of systems where, for each two pair numbers X and Y, if the winners from
the two pairs play each other in the next round, then the losers from these pairs also
play each other in the next round.
6.2. Theorem on neat systems (winners losers)
Premise: x S
1
Thesis: x S
2

6.3. Proof (indirect)
Figure 6 presents three different fragments of the system where moves from any
selected round (pairs X, Y, M) to the next round (pairs A, B, N) are marked. In the
first round (enclosed within a broken line), the pairs situated to the left are higher than
those to the right. According to the figure, the winners from pairs X and Y meet in the
Management of the teaching process

73
next round, but the losers from these pairs do not. Suppose the loser from pair X next
plays the loser from any other pair M. The examples presented below take into ac-
count the three possible locations of M with regard to bases X and Y. It can be seen
that each x such that x S
2
and x S
1
presented as a system of graphs has to include
at least one of the following three subgraphs.

Fig. 6. Three possible subgraphs describing a system x, such that x S
1
x S
2
It is enough to prove that a neat system x S
1
cannot contain any of the three pre-
sented fragments, i.e. x S
2
.
For the first subgraph of the system from Fig. 6, pair Y is higher than pair M
(Y > M), so, with regard to the neatness of the system, pair B should be placed higher
than pair A (B > A). At the same time, as pair X is higher than pair Y (X > Y), pair A
should be higher than pair B (A > B). We thus obtain a contradiction. This means that
the neat system x S
1
cannot contain the first of the three subgraphs. Similarly, one
can prove that the neat system x S
1
cannot contain the second of the three subgraphs.
It is also necessary to prove that an x S
1
neat system with the third subgraph is
not possible. Even if the subgraph is extended by adding extra pairs, it cannot contain
one that is isomorphic to either the first or the second subgraph (it must look like
a zigzag). Thus, the number of pairs in a previous round must always be smaller
than in a later one. This is impossible, as the number of pairs in each round is constant.
Hence, the losers of pairs X and Y have to meet in the next round, i.e. x S
2
.

Fig. 7. An example of a neat system y such that y S
1
= S
2

Thus, S
1
S
2
. Since one can prove that S
2
S
1
, then S
1
= S
2
.
There arises the question of whether there are any neat y systems such that
y S
1
= S
2
. If pair A can be joined by a winner from pair M and not a loser, as fol-
lows from the assumption of the theorem, then it is possible to create a system, e.g. the
R. SZOSTEK

74
one shown in Fig. 7. In this system, a winner meets a loser in pair A, so it is an exam-
ple of a neat system y such that y S
1
= S
2
. In the example above, each of the pairs:
N, X, Y, A, B, M is different, but, in general, the pairs X, Y, M may coincide with
some of the pairs A, B, N.
7. Conclusion
The paper presents an effective system for managing sports competitions. Such
problems can be formulated as one of probabilistic sorting. The proposed method of
organizing competitions is novel and has assets that other known and currently applied
methods of sports competition management do not possess:
1. A competition can consist of a practically unlimited number of teams, so there
is no need to play in leagues.
2. The number of sequences played in a season can be established in advance.
3. Each team plays the same number of matches in a season, as none of them are
eliminated from the competition.
4. Each team takes a specific, individual place in the ranking, owing to the fact
that pairs become more uniform after each round is completed. There are no groups
of death, which are unfair.
5. Teams compete with each other on various levels. On these levels, constant
competition concerning promotion to a higher level and defense against a fall to a
lower level is in progress. Thus, the games are a source of entertainment.
6. Strong teams play more frequently with other strong teams, whereas weak
teams compete more frequently with other weak teams. Therefore, there are less un-
necessary matches to play. Moreover, each team competes with teams of a similar
quality. Therefore, fewer matches that are unnecessary or uncompetitive are played in
general.
7. There are several big clashes in each round. Such matches are played between
top teams.
8. There is a chance that teams that have already played against each other will
meet again. This is possible if their quality and number of points obtained are similar.
In this way, only similar teams are compared on a relatively large number of occa-
sions.
9. It makes corruption difficult, as matches are played between teams that are clas-
sified as being of a similar level. Furthermore, the result of each match is likely to be
of great importance to the final ranking.
There are still numerous open questions. How is the probability of establishing the
true ranking affected by the following:
Management of the teaching process

75
the starting arrangement,
the variant of the algorithm presented,
the number of games to be played,
the form of the function from Fig. 1,
the number of levels on which comparisons take place,
the number of points obtained by each team.
The probability of establishing the true ranking regarding each of the above ques-
tions can be treated as the probability that the ordering of the teams made according to
the sorting algorithm coincides with the ordering of their qualities (a global order) or
as the probability that a specific team will be found in the appropriate position in the
final ranking (a local order).
References
[1] Illustrated Encyclopaedia of Sports, Muza S.A. Publ., Poland, 2001 (in Polish).
[2] HAREL D., Algorythmics: the Spirit of Computing, Pearson Education, 1992.
[3] MITZENMACHER MI., UPFAL E., Probability and Computing. Randomized Algorithms and Probabilis-
tic Analysis, Cambridge University Press, 2006.
[4] WIRTH N ., Algorithms + Data Structures = Programs, Prentice Hall, 1976.