=Paper=
{{Paper
|id=Vol-2332/paper-11-004
|storemode=property
|title=
Using the Gini coefficient to calculate the degree of consensus in group decision making process
|pdfUrl=https://ceur-ws.org/Vol-2332/paper-11-004.pdf
|volume=Vol-2332
|authors=Nadezhda V. Chukhno,Olga V. Chukhno,Irina A. Gudkova,Konstantin E. Samouylov
}}
==
Using the Gini coefficient to calculate the degree of consensus in group decision making process
==
https://ceur-ws.org/Vol-2332/paper-11-004.pdf
97
UDC 519.816
Using the Gini coefficient to calculate the degree of consensus
in group decision making process
Nadezhda V. Chukhno* , Olga V. Chukhno* ,
Irina A. Gudkova*† , Konstantin E. Samouylov*‡
*
Peoples’ Friendship University of Russia (RUDN University)
6 Miklukho-Maklaya St, Moscow, 117198, Russian Federation
†
Faculty of Electrical Engineering and Communication
Brno University of Technology
3058/10 Technická, 61600 Brno, Czech Republic
‡
Federal Research Center “Computer Science and Control” of the Russian Academy of
Sciences (FRC CSC RAS)
44-2 Vavilov St, Moscow, 119333, Russian Federation
Email: nadezdachukhno@yandex.ru, olga-chukhno95@yandex.ru,gudkova-ia@rudn.ru, samuylov-ke@rudn.ru
Nowadays, almost all people in the world can communicate and exchange information on
the Internet. The Internet is an open structure where everyone can express their opinions.
In social networks millions of users are registered around the world, and the data of their
interaction can be used for specific purposes, for instance, in group decision making problems.
Therefore, social network is one of the best environments to raise any questions, discuss them
and make decisions. The process of reaching consensus and many different approaches to
the solution of this problem have already been well studied. In recent years, modeling of
the process of reaching consensus in the context of social networks is of special interest. In
addition, the development of improved structures for GDM processes and consensus decision
making is actual now, as they can be used in new social networking services. Group decision
making is the process of selecting the best alternative or a set of alternatives from all possible.
In conditions that are far from reality, decision-makers come to full agreement. However, most
often such result is impossible. Actually, it is interesting to understand how much the experts
had reached agreement through discussion. Calculation of the degree of agreement usually
requires the calculation and aggregation of distance measures, which assess how close each
expert’s preferences to each pair of alternatives. Such calculations can have long time in view
of the selected aggregation operator and require the construction of a collective preference
matrix before it can be obtained. In this paper we propose to use the Gini coefficient and
present the formula, which shows the degree of agreement between the experts. We also offer
a method for assessing the effectiveness of consensus reached by decision-makers.
Key words and phrases: group decision making, social network analysis, fuzzy logic,
Gini coefficient, consensus level.
Copyright © 2018 for the individual papers by the papers’ authors. Use permitted under the CC-BY license —
https://creativecommons.org/licenses/by/4.0/. This volume is published and copyrighted by its editors.
In: K. E. Samouylov, L. A. Sevastianov, D. S. Kulyabov (eds.): Selected Papers of the 12th International Workshop on
Applied Problems in Theory of Probabilities and Mathematical Statistics (Summer Session) in the framework of the
Conference “Information and Telecommunication Technologies and Mathematical Modeling of High-Tech Systems”,
Lisbon, Portugal, October 22–27, 2018, published at http://ceur-ws.org
98 APTP+MS’2018
1. Introduction
Social network analysis is a new and actively developing direction in the field of
group decision making (GDM). The interest of researchers in this field is due to the fact
that it provides a set of explanatory models and analytical tools beyond the scope of
conventional quantitative methods. At the same time, a rich mathematical apparatus
has been accumulated, which makes it possible to build complex models of social
interactions [1].
One of the advantages that the Internet provides is the opportunity for everyone to
express their opinions, which allows making group decisions based on the information
gathered. It should be noted that social network is the most convenient platform for
achieving consensus decisions [2].
Group decision making is applied in many situations in the real world. Firstly, GDM
is a process of ranking of alternatives, taking into account the opinions expressed by a
group of people [3, 4]. Secondly, it is a process focused on people, with their subjectivity
and uncertainty in evaluation. Therefore, in GDM processes commonly fuzzy logic
theory is used [5, 6]. When we use the apparatus of fuzzy logic, alternatives are classified
according to the following principle: the degree of preference is expressed as a value
from the interval [0,1] with 0 for less preferable and 1 for more preferable.
Group decision making problems require a high level of consensus among experts
before getting a solution. A prospective field of research of group decision making is the
study of interpersonal interaction and calculation the degree of consensus agreement [7,8].
The main goal of our work is to evaluate the effectiveness of the consensus reached.
The rest of the paper is organized as follows: in section 2 a model of group decision
making in a social network is formally described. Section 3 demonstrates the new
coefficient to calculare the degree of agreement among all experts.In section 4 we present
the results of numericai example of applying this method. Section 5 concludes the paper.
2. The model of consensus in group decision making process
The goal of the group decision making process is to order different alternatives from
a set of alternatives from the best to the worst with the help of the association of some
preference degrees, taking into account the opinions of groups of decision-makers [9, 10].
We assume that the GDM process involves the participation of K experts who compare
M alternatives pairwise. Then 𝐸 = {𝑒1 , ..., 𝑒𝐾 } - a set of experts and 𝑋 = {𝑥1 , ..., 𝑥𝑀 }
– a set of alternatives, such that |𝐸| = 𝐾 < ∞, where K is the number of experts and
|𝑋| = 𝑀 < ∞, where M – the number of alternatives.
Figure 1. The scheme of GDM process.
Each expert 𝑒𝑘 provides a certain preference value 𝑝𝑖𝑗 (𝑘), which indicates how much
the alternative 𝑖 is better than the alternative 𝑗.
Chukhno N. V. et al. 99
The obtained values form the preference matrix 𝑃𝑘 = (𝑝𝑖𝑗 (𝑘)),𝑖,𝑗∈1,...,𝑀 for the
expert 𝑒𝑘 , where 𝑝𝑖𝑗 = 1 reflects the maximum preference, 𝑝𝑖𝑗 = 0 - the minimum
preference. And 𝑝𝑖𝑗 = 1 − 𝑝𝑗𝑖 , 𝑝𝑖𝑖 = 0.5.
The GDM process contains following stages.
1. Providing preferences. Experts decide which alternatives are most appropriate
and share their opinions to the system. One of the most commonly used techniques
for implementing this process is to allow users to compare alternatives pairwise.
2. Aggregation of information received from all experts. Individual preferences
of experts are aggregated into the matrix of collective preferences. The selected
aggregation operator summarizes the preferences or reflects properties contained in
the preferences of experts. In this stage we prefer to use the matrix of averaged
estimate 𝑃 =(𝑝𝑖𝑗 )𝑖,𝑗=1,...,𝑀 , the elements of which are calculated as follows
𝐾
∑︀
𝑝𝑖𝑗 (𝑘)
𝑘=1
𝑝𝑖𝑗 = . (1)
𝐾
Also OWA operators can be used for this purpose. The obtained result contains
the general opinion of all experts. It is usually presented in the form of a square
matrix, where each position 𝑝𝑖𝑗 contains the preference of the alternative 𝑥𝑖 over
𝑥𝑗 .
3. Exploitation of the information received from stage 2. This step forms the
final ranking of alternatives. Results can be presented as a choice of the best
alternative or as a rating of all alternatives.
We can perform the ranking of alternatives using, for instance, operators GDD
(Quantifier Guided Dominance Degree) and GNDD (Quantifier Guided Non-Dominance
Degree). These operators are aimed to calculate the rating of alternatives or create lists
of ranked alternatives using the collective preference matrix obtained in the previous
step. Commonly alternatives are evaluated using an average value between GDD and
GNDD. The operator GDD shows how the estimated alternative dominates all the
others, i.e. how much this alternative is better than all the others.
The calculation of GDD is carried out according to the following formula
𝑀
∑︁
𝐺𝐷𝐷𝑖 = 𝑝𝑖𝑗 , 𝑖 = 1, ..., 𝑀 . (2)
𝑗=1
The GNDD operator can be calculated as follows
𝑀
∑︁
𝐺𝑁 𝐷𝐷𝑖 = 1 − max{𝑝𝑗𝑖 − 𝑝𝑖𝑗 , 0}, 𝑖 = 1, ..., 𝑀 . (3)
𝑗=1
Averaging the values of GDD and GNDD, we obtain the RV (Ranking Value):
𝐺𝐷𝐷𝑖 + 𝐺𝑁 𝐷𝐷𝑖
𝑅𝑉𝑖 = , 𝑖 = 1, ..., 𝑀 . (4)
2
When a decision is defined as a value of 𝑅𝑉𝑖 for each alternative (4), it is interesting
to understand how much each of the experts agree with its, in other words, to know the
degree of agreement among the experts.
4)Calculating the consensus level among experts in a group decision making
process is an important part of this process. Thanks to this, experts know whether they
have reached agreement or, on the contrary, their opinions are too far from each other.
Therefore, consensus measures help to decide whether to continue the discussion or they
have already reached an agreement.
100 APTP+MS’2018
Figure 2. Group decision making model.
3. The Gini coefficient for calculating the degree of agreement among
experts
In most real-world scenarios, a complete consensus is practically unreachable. Due
to some differences in the level of knowledge and personal interests of decision-makers, a
full agreement is reached in rare cases. To understand how much experts were close to
each other in expression opinions on the set of alternatives, we propose to use the Gini
coefficient.
Traditionally, this coefficient is used in the economy and social policy for differentiation
of incomes of the population. It is necessary to find the value at which the distribution
of the economic variable deviates from the ideal value (equal distribution of wealth at
all people). For measuring a certain statistical spread or dispersion are used.
The degree of agreement (consensus level) among the participants of the group
decision making process is one of the most important indicators. The measurement of
this coefficient is a necessary task before making a final decision.
In GDM problems, the measurement of the degree of agreement determines as well
as a high degree of similarity in the distribution of preference values.
The Lorenz curve in the economy makes it possible to define the degree of income
inequality population. So it can also demonstrate unequal distribution in any system.
It should be stressed that the Gini coefficient is intimately connected with the Lorentz
curve and is equal to the ratio of the area of the figure bounded by the line of absolute
equality and the Lorentz curve, to the area of the entire triangle under the line of
absolute equality. Define 𝐴 - area of the region between line of absolute equality and
the curve, 𝐵 - area of the region under the curve, 𝐶 - area of the region above the curve
as shown in Fig. 3:
𝐴
𝐼𝐺𝑖𝑗 = . (5)
𝐴+𝐵
Theorem 1.
Theorem 1. Gini consensus coefficient for each pair of alternatives (i,j) is defined as
𝐾−1
2 ∑︁
𝐼𝐺𝑖𝑗 = 1 − ℎ𝑖𝑗 (𝑘). (6)
𝐾 𝑘=1
Proof.
Chukhno N. V. et al. 101
Figure 3. Lorenz Curve.
𝑘
∑︀
Proof. Define ℎ𝑖𝑗 (𝑘) = 𝑞𝑖𝑗 (𝑙) , where
𝑙=1
𝐾
⎧
𝑝𝜎
∑︀
⎨0, if 𝑖𝑗 (𝑙) = 0,
⎪
⎪
⎪
𝑙=1
𝑞𝑖𝑗 (𝑘) = 𝑝𝜎 (𝑘)
𝑖𝑗
(7)
⎪
⎪
⎪ 𝐾
, elsewhere,
⎩ ∑︀ 𝑝𝑖𝑗 (𝑙)
𝑙=1
Then by the definition [7] and formula (7) the Gini coefficient corresponds to the
formula (5), where 𝐶 + 𝐵 = 1, 𝐴 + 𝐵 = 𝐶 − 𝐴 = 0.5.
From equalities
𝐾 𝐾 𝐾 𝐾−1
∑︁ 𝑘 1 ∑︁ 1 ∑︁ 1 ∑︁
𝐶= = 𝑘𝑞𝑖𝑗 (𝑘) = 𝑘(ℎ𝑖𝑗 (𝑘) − ℎ𝑖𝑗 (𝑘 − 1)) = 1 − ℎ𝑖𝑗 (𝑘)
𝑘=1
𝐾𝑞𝑖𝑗 (𝑘) 𝐾 𝑘=1 𝐾 𝑘=1 𝐾 𝑘=1
we get that
𝐾 𝐾−1
𝐶 − 0.5 2 ∑︁ 2 ∑︁
𝐼𝐺𝑖𝑗 = = 𝑘𝑞𝑖𝑗 (𝑘) − 1 = 1 − ℎ𝑖𝑗 (𝑘). (8)
0.5 𝐾 𝑘=1 𝐾 𝑘=1
𝐼𝐺𝑖𝑗 = 0 shows the perfect agreement, where all evaluations of experts are the same,
and 𝐼𝐺𝑖𝑗 = 1 represents the maximum difference between evaluations.
102 APTP+MS’2018
Table 1. Set of alternatives.
𝑥𝑖 Mobile OS
𝑥1 Tizen
𝑥2 Windows 10 Mobile
𝑥3 Lineage OS
𝑥4 Android
𝑥5 iOS
𝑥6 Fire OS
𝑥7 Kai OS
4. Case study
In this section we give a numerical example of the proposed coefficient.
Let us consider such process, in which 5 experts 𝐸 = {𝑒1 , 𝑒2 , 𝑒3 , 𝑒4 , 𝑒5 } participate
and there are 7 alternatives 𝑋 = {𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 , 𝑥5 , 𝑥6 , 𝑥7 }. We have conducted a survey
to determine which of the 7 below listed mobile operating systems are most convenient
for using according to 5 experts.
Thus, taking into account the notation introduced earlier, for the analysis we have
the initial data presented in Table 1.
Experts’ preferences are presented in the following matrixes
⎛ ⎞
0.5 0.4 0.8 0.5 0.2 0.7 0.7
⎜0.6 0.5 1 0.5 0.3 0.8 0.5⎟
⎜ ⎟
⎜ ⎟
⎜0.2 0 0.5 0.1 0 0.5 0.2⎟
⎜ ⎟
𝑃1 = ⎜⎜0.5 0.5 0.9 0.5 0.4 0.9 0.8⎟
⎟
⎜0.8 0.7 1 0.6 0.5 1 0.7⎟
⎜ ⎟
⎜ ⎟
⎝0.3 0.2 0.5 0.1 0 0.5 0.4⎠
0.3 0.5 0.8 0.2 0.3 0.6 0.5
,
⎛ ⎞
0.5 0.3 0.8 0.5 0.1 0.5 0.5
⎜0.7 0.5 0.7 0.4 0.3 0.5 0.6⎟
⎜ ⎟
⎜ ⎟
⎜0.2 0.3 0.5 0.3 0.2 0.5 0.4⎟
⎜ ⎟
𝑃2 = ⎜
⎜0.5 0.6 0.7 0.5 0.5 0.8 0.7⎟
⎟
⎜0.9 0.7 0.8 0.5 0.5 8 0.9⎟
⎜ ⎟
⎜ ⎟
⎝0.5 0.5 0.5 0.2 0.2 0.5 0.5⎠
0.5 0.4 0.6 0.3 0.1 0.5 0.5
,
...
Chukhno N. V. et al. 103
⎛ ⎞
0.5 0.6 0.6 0.4 1 0.6 0.7
⎜0.4 0.5 0.8 0.5 0.2 0.7 0.5⎟
⎜ ⎟
⎜ ⎟
⎜0.4 0.2 0.5 0.2 0.1 0.5 0.5⎟
⎜ ⎟
𝑃5 = ⎜
⎜0.6 0.5 0.8 0.5 0.6 0.8 0.9⎟
⎟
⎜ 0 0.8 0.9 0.4 0.5 0.6 0.9⎟
⎜ ⎟
⎜ ⎟
⎝0.4 0.3 0.5 0.2 0.4 0.5 0.7⎠
0.3 0.5 0.5 0.1 0.1 0.3 0.5
,
which result in the following matrix of averaged estimates, calculated by (1):
⎛ ⎞
0.5 0.48 0.66 0.38 0.52 0.66 0.58
⎜0.52 0.5 0.72 0.42 0.3 0.6 0.58⎟
⎜ ⎟
⎜ ⎟
⎜0.34 0.28 0.5 0.24 0.26 0.44 0.44⎟
⎜ ⎟
𝑃 =⎜⎜0.62 0.58 0.76 0.5 0.5 0.7 0.84⎟⎟
⎜0.48 0.7 0.74 0.5 0.5 0.68 0.8 ⎟
⎜ ⎟
⎜ ⎟
⎝0.34 0.4 0.56 0.3 0.32 0.5 0.64⎠
0.42 0.42 0.56 0.16 0.2 0.36 0.5
.
Then using GDD operator (2) and GNDD (3), we get the following results
𝐺𝐷𝐷1 = 3.78;
𝐺𝐷𝐷2 = 3.64;
𝐺𝐷𝐷3 = 2.5;
𝐺𝐷𝐷4 = 4.5;
𝐺𝐷𝐷5 = 4.5;
𝐺𝐷𝐷6 = 3.06;
𝐺𝐷𝐷7 = 2.62;
𝐺𝑁 𝐷𝐷1 = 5.72;
𝐺𝑁 𝐷𝐷2 = 6.44;
𝐺𝑁 𝐷𝐷3 = 5;
𝐺𝑁 𝐷𝐷4 = 7;
𝐺𝑁 𝐷𝐷5 = 6.96;
𝐺𝑁 𝐷𝐷6 = 5.72;
𝐺𝑁 𝐷𝐷7 = 5.12.
The ranking of alternatives is given by (4):
𝑅𝑉1 = 3.78+5.72
2
= 4.75;
𝑅𝑉2 = 3.64+6.44
2
= 5.04;
𝑅𝑉3 = 2.5+5
2
= 3.75;
𝑅𝑉4 = 4.5+7
2
= 5.75;
𝑅𝑉5 = 4.5+6.96
2
= 5.68;
𝑅𝑉6 = 3.06+5.72
2
= 4.39;
𝑅𝑉7 = 2.62+5.12
2
= 3.87.
So, we get the following ranking list: 𝑥4 , 𝑥5 , 𝑥2 , 𝑥1 , 𝑥6 , 𝑥7 , 𝑥3 .
According to this list, the best mobile operating systems in terms of convenience are
Android, IOS and Windows.
Then applying the formula (6) or (8) to the preference matrixes, we have
104 APTP+MS’2018
Figure 4. The expert preferences in mobile operating systems.
⎛ ⎞
0.5 0.37 0.3 0.37 0.58 0.28 0.31
⎜0.35 0.5 0.33 0.33 0.31 0.31 0.27⎟
⎜ ⎟
⎜ ⎟
⎜0.39 0.54 0.5 0.54 0.72 0.31 0.36⎟
⎜ ⎟
𝐼𝐺 = ⎜
⎜ 0.3 0.3 0.31 0.5 0.27 0.32 0.25⎟
⎟
⎜0.62 0.25 0.38 0.27 0.5 0.37 0.26⎟
⎜ ⎟
⎜ ⎟
⎝0.36 0.36 0.29 0.5 0.55 0.5 0.38⎠
0.35 0.3 0.33 0.45 0.44 0.5 0.5
.
Considering that it is necessary to obtain a degree of expert consensus on all alterna-
tives, we get
𝑀 ∑︀
∑︀ 𝑀
𝐼𝐺𝑖𝑗
𝑖=1 𝑗=1
𝐼𝐺 = , 𝑖 ̸= 𝑗.
𝑀
∑︀
2 (𝑀 − 𝑟)
𝑟=1
Thus, the average degree of agreement is 𝐼𝐺 = 0.37.
The Gini coefficient indicates that the experts have reached a high level consensus,
because of this coefficient is closer to 0 than to 1. Therefore, we can conclude that the
experts’ estimates are near equal to each other.
Chukhno N. V. et al. 105
5. Conclusions
In this paper a new consensus coefficient proposed in [7, 8] for GDM problems with
fuzzy preferences relations was studied. We have investigated the process of group
decision making and analyzed the consensus degree to which the experts agree with
the decision. For that we built a model of consensus in group decision making, made a
rating of alternatives, got a formula for measuring the degree of agreement of experts
involved in the GDM process. Also we have proved a formula for measuring the degree
of experts’ agreement and calculated the consensus level.
Results showed consistent and acceptable behavior of the proposed coefficient, which
justifies its use as a valid consensus measure to solve the problems of group decision
making.
It should be noted that the Gini coefficient has some interesting advantages. Firstly,
it is possible to calculate the degree of consensus without using distances measures. In
this case, the use of this coefficient will replace the time-consuming calculations that can
not be avoided in the classical theory [1]. Secondly, the Gini does not depend on various
aggregation operators that can be applied to obtain a collective preference matrix. And
also, the value of the coefficient will not change, even if the final decision changes. This is
a kind of evolution of the degree of agreement, independent of the aggregation operator.
Acknowledgments
The publication has been prepared with the support of the “RUDN University
Program 5-100” and funded by RFBR according to the research projects No. 17-07-
00845, 18-00-01555.
References
1. Wasserman, S., Faust, K. Social Network Analysis: Methods and Applications (Vol.
8) // Cambridge: Cambridge University Press. – 1994.
2. Capuano, N., Chiclana, F., Fujita, H., Herrera-Viedma, E., Loia, V. Fuzzy Group
Decision Making With Incomplete Information Guided by Social Influence// IEEE
Transactions on Fuzzy Systems, 26(3), 1704–1718, 2018.
3. Herrera-Viedma, E., Cabrerizo, F., Chiclana, F., Wu, J., Cobo, M., Samuylov, K.
Consensus in group decision making and social network// Studies in informatics
and control, issn 1220-1766. Vol. 26(3). P. 259-268, 2017.
4. Herrera-Viedma, E., Cabrerizo, F.J., Kacprzyk, J., Pedrycz, W. A review of soft
consensus models in a fuzzy environment // Information Fusion. 17. P. 4-13, 2014.
5. Tanino, T. Fuzzy preference orderings in group decision making // Fuzzy Sets and
Systems. 12. P. 117–131, 1984.
6. Orlovsky, S.A. Decision-Making with a Fuzzy Preference Relation // Fuzzy Sets
and Systems. P. 155-167, 1978.
7. Haughton, J., Khandker, S.R. // Handbook on Poverty and Inequality, Washington
DC, World Bank, 2009.
8. Del Moral,M.J., Chiclana, F., Tapia, J.M., Herrera-Viedma, E. A comparative
study on consensus measures in group decision making // International Journal of
Intelligent Systems 1-15, 2018.
9. Samuylov, A., Moltchanov, D., Gaidamaka, Y., Andreev, S., Koucheryavy, Y.
Random Triangle: A Baseline Model for Interference Analysis in Heterogeneous
Networks // IEEE Transactions on Vehicular Technology, 65 (8), art. no. 7275184,
pp. 6778-6782, 2016.
10. Begishev, V., Kovalchukov, R., Samuylov, A., Ometov, A., Moltchanov, D.,
Gaidamaka, Y., Andreev, S. An analytical approach to SINR estimation in adjacent
rectangular cells // Lecture Notes in Computer Science (including subseries Lecture
Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 9247, pp.
446-458, 2015.