Skip to main content
Log in

Evaluating groups with the generalized Shapley value

  • Research Paper
  • Published:
4OR Aims and scope Submit manuscript

Abstract

Following the original interpretation of the Shapley value as a priori evaluation of the prospects of a player in a multi-person interaction situation, we intend to apply the Shapley generalized value (introduced formally in Marichal et al. in Discrete Appl Math 155:26–43, 2007) as a tool for the assessment of a group of players that act as a unit in a coalitional game. We propose an alternative axiomatic characterization which does not use a direct formulation of the classical efficiency property. Relying on this valuation, we also analyze the profitability of a group. We motivate this use of the Shapley generalized value by means of two relevant applications in which it is used as an objective function by a decision maker who is trying to identify an optimal group of agents in a framework in which agents interact and the attained benefit can be modeled by means of a transferable utility game.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3

Image taken from Lindelauf et al. (2013)

Fig. 4

Image taken from V.E. Krebs (Copyright \(\copyright \)2002, First Monday)

Fig. 5

Image taken from V.E. Krebs (Copyright \(\copyright \)2002, First Monday)

Similar content being viewed by others

Notes

  1. We will use interchangeably the two terminologies, game and characteristic function, when no ambiguity appears.

  2. But not, as said above, the problem of assigning values to groups.

  3. They have been estimated by means of a Monte Carlo simulation following Castro et al. (2009) of the value of every group obtained with 7000 replications of the experiment.

  4. This fact will be analyzed in next section.

  5. Actually, Derks and Tijs (2000) refer to profitability as mergeability.

  6. Note that the kind of group integration we work with is equivalent to the collusive contracts considered by Segal (2003).

  7. The authors collected the strength of existing relationships from Koschade (2006).

  8. In which we have again considered the \(w^{conn2}\) game, with a zero weight for all the terrorists who do not take direct part in the attacks.

References

  • Amer R, Gimenez JM (2004) A connectivity game for graphs. Math Methods Oper Res 60:453–470

    Article  Google Scholar 

  • Aumann RJ, Drèze JH (1974) Cooperative games with coalition structures. Int J Game Theory 3:217–237

    Article  Google Scholar 

  • Bulow JI, Geanakoplos JD, Klemperer PD (1985) Multimarket oligopoly: strategic substitutes and complements. J Polit Econ 93:488–511

    Article  Google Scholar 

  • Castro J, Gomez D, Tejada J (2009) Polynomial calculation of the Shapley value based on sampling. Comput Oper Res 36:1726–1730

    Article  Google Scholar 

  • Crama Y, Leruth L (2013) Power indices and the measurement of control in corporate structures. Int Game Theory Rev 15(3):1340017

    Article  Google Scholar 

  • Derks J, Tijs S (2000) On merge properties of the Shapley value. Int Game Theory Rev 2:249–257

    Article  Google Scholar 

  • Flores R, Molina E, Tejada J (2014) The Shapley group value. arXiv:1412.5429 [math.OC]

  • Fujimoto K, Kojadinovic I, Marichal JL (2006) Axiomatic characterizations of probabilistic and cardinal-probabilistic interaction indices. Games Econ Behav 55:72–99

    Article  Google Scholar 

  • Grabisch M, Roubens M (1999) An axiomatic approach to the concept of interaction among players in cooperative games. Int J Game Theory 28:547–565

    Article  Google Scholar 

  • Granovetter M (1978) Threshold models of collective behavior. Am J Sociol 83:1420–1443

    Article  Google Scholar 

  • Gul F (1989) Bargaining foundations of Shapley value. Econometrica 57:81–95

    Article  Google Scholar 

  • Hamlen SS, Hamlen WA Jr, Tschirhart J (1980) The use of the generalized Shapley allocation in joint cost allocation. Account Rev 55:269–287

    Google Scholar 

  • Hart S, Kurz M (1983) Endogenous formation of coalitions. Econometrica 51:1047–1064

    Article  Google Scholar 

  • Kalai E, Samet D (1987) On weighted Shapley values. Int J Game Thoery 16:205–222

    Article  Google Scholar 

  • Kempe D, Kleinberg J, Tardos E (2005) Influential nodes in a diffusion model for social networks. In: Proceedings of the 32nd international colloquium on automata, languages and programming, pp 1127–1138

  • Koschade S (2006) A social network analysis of Jeemaah Islamiyah: the applications to counterterrorism and intelligence. Stud Confl Terror 29:559–575

    Article  Google Scholar 

  • Krebs V (2002) Mapping networks of terrorist cells. Connections 24:43–52

    Google Scholar 

  • Lehrer E (1988) An axiomatization of the Banzhaf value. Int J Game Theory 17:89–99

    Article  Google Scholar 

  • Lindelauf R, Hamers HJM, Husslage BGM (2013) Cooperative game theoretic centrality analysis of terrorist networks: the cases of Jemaah Islamiyah and Al Qaeda. Eur J Oper Res 229:230–238

    Article  Google Scholar 

  • Liu F, Fan Y, Ordoñez F (2009) A two-stage stochastic programming model for transportation network protection. Comput Oper Res 36:1582–1590

    Article  Google Scholar 

  • Lucchetti R, Moretti S, Patrone F (2015) Ranking sets of interacting objects via semivalues. TOP 23:567–590

    Article  Google Scholar 

  • Marichal JL (2000) The influence of variables on pseudo-Boolean functions with applications to game theory and multicriteria decision making. Discrete Appl Math 107:139–164

    Article  Google Scholar 

  • Marichal JL, Kojadinovic I, Fujimoto J (2007) Axiomatic characterizations of generalized values. Discrete Appl Math 155:26–43

    Article  Google Scholar 

  • McQuillin B (2009) The extended and generalized Shapley value: simultaneous consideration of coalitional externalities and coalitional structure. J Econ Theory 144:696–721

    Article  Google Scholar 

  • Moretti S, Tsoukias A (2012) Ranking sets of possibly interacting objects using Shapley extensions. In: Thirteenth international conference on the principles of knowledge representation and reasoning (KR 2012). AAAI Press, pp 199–209

  • Myerson RB (1977) Graphs and cooperation in games. Math Oper Res 2:225–229

    Article  Google Scholar 

  • Narayanam R, Narahari Y (2011) A shapley value-based approach to discover influential nodes in social networks. IEEE Trans Autom Sci Eng 99:1–18

    Google Scholar 

  • Owen G (1977) Values of games with a priori unions. In: Henn R, Moeschlin O (eds) Essays in mathematical economics and game theory. Springer, New York

    Google Scholar 

  • Segal I (2003) Collusion, exclusion and inclusion in random-order bargaining. Rev Econ Stud 70:439–460

    Article  Google Scholar 

  • Shapley LS (1953a) A value for \(n\)-person games. In: Kuhn HW, Tucker AW (eds) Contributions to the theory of games, vol II. Princeton University Press, Princeton, pp 307–317

    Google Scholar 

  • Shapley LS (1953b) Additive and non-additive set functions. Ph.D. thesis, Princeton University

  • Vitali S, Glattfelder JB, Battiston B (2011) The network of global corporate control. PLOS ONE 6:e25995

    Article  Google Scholar 

  • Winter E (2002) The Shapley value. In: Aumann RJ, Hart S (eds) Handbook of game theory with economic applications, vol 3. North-Holland Publishing Company, Amsterdam, pp 2025–2054

    Google Scholar 

Download references

Acknowledgements

We would like to warmly thank Javier Castro (Universidad Complutense de Madrid) for the simulation program used to obtain the numerical results in Sect. 5.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Ramón Flores.

Additional information

This research has been supported by I+D+i research projects MTM2011-27892 and MTM2015-70550 from the Government of Spain.

Appendix

Appendix

Theorem 1

The unique group value over the set of all games \({\mathcal {G}}\) verifying G-null player, G-linearity, G-CBC, and G-SPB is the Shapley group value \(\phi ^g\).

Proof

Let us first prove that the Shapley group value \(\phi ^g\) satisfies the previous four properties.

Let us arbitrarily fix the sets C and N, \(C\subset N\in {\mathcal {N}}\), and let \(v\in {\mathcal {G}}_N\).

To check properties P1 and P2 the reader is referred to Marichal et al. (2007).

With respect to property P3, G-CBC, let \(C\subset N\in {\mathcal {N}}\) be any two finite sets, and let v any game in \({\mathcal {G}}_N\). First let us remark that condition (3) is equivalent to:

$$\begin{aligned}&\xi ^g({C\cup i};N,v)-\xi ^g({C\cup j};N,v) \nonumber \\&\quad =(\xi ^g({C\cup i};N\backslash j,v_{-j})-\xi ^g(C;N\backslash j,v_{-j}))\nonumber \\&\qquad -(\xi ^g({C\cup j};N\backslash i,v_{-i})-\xi ^g(C;N\backslash i,v_{-i})). \end{aligned}$$
(10)

Also note that, by definition of the merging game and the Shapley value, for any \(i,j\in N{\setminus } C\) the following equalities hold:

$$\begin{aligned} \displaystyle \phi ^g({C\cup i};N,v)= & {} \displaystyle \sum _{\begin{array}{c} S\subset N{\setminus } C \\ i,j\notin S \end{array}} \Bigl ( \frac{s!(n-c-s-1)!}{(n-c)!}(v(S\cup C\cup i)-v(S))\\&+\frac{(s+1)!(n-c-s-2)!}{(n-c)!}(v(S\cup C\cup i\cup j)-v(S\cup j))\Bigr ), \\ \phi ^g(C;N\backslash j,v_{-j})= & {} \displaystyle \sum _{\begin{array}{c} S\subset N{\setminus } C\\ i,j\notin S \end{array}} \Bigl (\frac{s!(n-c-s-1)!}{(n-c)!}(v(S\cup C)-v(S))\\&+\frac{(s+1)!(n-c-s-2)!}{(n-c)!}(v(S\cup C\cup i)-v(S\cup i))\Bigr ), \\ \phi ^g ({C\cup j};N\backslash i,v_{-i})= & {} \sum _{\begin{array}{c} S\subset N{\setminus } C\\ i,j\notin S \end{array}} \frac{s!(n-c-s-2)!}{(n-c-1)!}(v(S\cup C\cup j)-v(S)). \end{aligned}$$

Analogous expressions hold for \(\phi ^g({C\cup j};N,v)\), \(\phi ^g (C;N{\setminus } i,v_{-i})\) and \(\phi ^g ({C\cup i}; N{\setminus } j, v_{-j})\). Now it is enough to check that for every \(S\in N\) with \(i,j\notin S\) the coefficients of \(v(S\cup C\cup i\cup j)\), \(v(S\cup C\cup i)\), \(v(S\cup C\cup j)\), \(v(S\cup C)\), \(v(S\cup i)\), \(v(S\cup j)\) and v(S) are the same in both sides of the equation in (10), and this is easily deduced from the previous expressions. We leave the details to the reader.

It remains property P4, G-SPB. Consider the unanimity game with respect to the grand coalition \((N,u_N)\), and a non-empty group \(C\subset N\in {\mathcal {N}}\). It is straightforward to see that the merging game \((N_C,(u_N)_C)\) is the unanimity game \((N_C,u_{N_C})\) with respect to the grand coalition \(N_C\), so \(\phi ^g (C;N,u_N):=\phi _{\mathbf {c}}(N_C,(u_N)_C)=\frac{1}{n-c+1}\), as desired. \(\square \)

Proof

We have proved that the properties hold for the Shapley group value, so we are left with the question of uniqueness.

Since \(\{(N,u_S)\}_{\begin{array}{c} S\subset N \\ S\ne \varnothing \end{array}}\) forms a basis of \({\mathcal {G}}_N\) for all \(N\in {\mathcal {N}}\), by G-linearity it is sufficient to consider the games \((N,u_S)\), \(\varnothing \ne S\subset N\in {\mathcal {N}}\). So let us see that \(\xi ^g (C;N,u_S)=\phi ^g (C;N,u_S):=\phi _{\mathbf {c}} (N_C,(u_{S})_C)\) for all non-empty subsets \(C,S\subset N\in {\mathcal {N}}\). When \(C=\varnothing \), this equality trivially holds by definition of a group value.

The proof will consist in a double induction over the cardinality of the player set N (first induction) and the cardinality of the unanimous coalition \(S\subset N\) (second induction).

First, we will prove that \(\xi ^g (C;N,u_S)=\phi ^g (C;N,u_S)\) for all non-empty subsets \(C,S\subset N\) whenever the cardinality of \(N\in {\mathcal {N}}\) is \(n\le 2\). For the unanimity game \((\{i\},u_i)\) with just one player \(i\equiv N\), G-SPB property P4 implies that \(\xi ^g (i;N,u_i)=1=\phi ^g(i;N,u_i)\). Now, let \((N,u_S)\) be a two-person unanimity game with \(N=\{i,j\}\). For the unanimity game \(u_N\), P4 implies that \(\xi ^g ( \{i,j\};N,u_N)=\phi ^g (\{i,j\};N,u_N)\). For the unanimity game \((N,u_i)\), G-null player P1 implies:

$$\begin{aligned} \xi ^g (j;N,u_i)=\xi ^g (\varnothing ;N,u_i):=0 \;\text { and } \; \xi ^g (i;N,u_i)=\xi ^g (\{i,j\};N,u_i) , \end{aligned}$$
(11)

Now, G-CBC property P3 implies

$$\begin{aligned} \xi ^g (i;N,u_i)-\xi ^g (j;N,u_i)&=(\xi ^g (i;\{i \},u_i)-\xi ^g({\varnothing };\{i \},u_i)-(\xi ^g(j;\{j \}, u^0) \\&\quad -\xi ^g({\varnothing };\{j \}, u^0)),= 1-0-0+0, \end{aligned}$$

where \(\xi ^g(i;N,u_i)=0\) for the trivial game \((\{i\},u^0)\) with \(u^0(i)=0\), follows from P1. Therefore, taking into account (11), \(\xi ^g (\cdot ;\{i,j\}, u_i)\equiv \phi ^g (\cdot ;\{i,j\}, u_i)\) holds. The same reasoning applies to \((N,u_j)\).

Let us fix a player set \(N\in {\mathcal {N}}\) with \(\vert N\vert =r\), and consider the unanimity game \((N,u_S)\) for a fix set \(\varnothing \ne S\subset N\). We will prove that \(\xi ^g (C;N,u_S)=\phi ^g (C;N,u_S)\) for all \(C\subset N\). Two cases are possible:

(i):

If \(S=N\), then G-SPB implies \(\xi ^g (C;N,u_N)=\phi ^g (C;N,u_N)\) for any non-empty group C in N.

(ii):

Otherwise, if \(\varnothing \ne S\subsetneq N\), we proceed by induction on the cardinality of C. Let us first prove the individual case \(C=\{i\}\).

There is at least a player \(j\in N\backslash S\) which by definition of unanimity game must be null. Let i be a player in S. Again by G-CBC, taking \(C=\varnothing \), we obtain \(\xi ^g (i;N,u_S)= \xi ^g (i;N\backslash j,{u_S} \vert _{N{\setminus } j})\), since G-null player implies \(\xi ^g (j;N,u_S)=\xi ^g ({\varnothing };N,u_S)=0\), and taking into account that \(u_S\vert _{N{\setminus } i}\equiv u^0\in {\mathcal {G}}_{N{\setminus } i}\).

Now, we may assume by the first induction that for every unanimity game \((N',u_S\vert _{N'})\) with \(N'\subsetneq N\) we have \(\xi ^g (C;N',u_S\vert _{N'})=\phi ^g (C;N',u_S\vert _{N'})\) for any group C in \(N'\). Thus, \(\xi ^g (i;N\backslash j,{u_S} \vert _{N{\setminus } j})= \phi ^g (i;N\backslash j,{u_S} \vert _{N{\setminus } j})\), which in turn is equal to \(\frac{1}{s}\) by definition, and then \(\xi ^g (i;N,u_S) =\frac{1}{s}=\phi _i(N,u_S)=\phi ^g(i;N,u_S)\). Note that every \(i\notin S\) is a null player in \(u_S\) and therefore G-null player implies \(\xi ^g (i;N,u_S)=0=\phi ^g (i;N,u_S)\). So we are done with the individual case \(C=\{i\}\).

Now, in order to prove \(\xi ^g (C;N,u_S)=\phi ^g (C;N,u_S)\) for all groups C with \(c>1\), we proceed by induction on the cardinality of C (second induction). So we take now \(1<r'\le r\), and we may assume that \(\xi ^g (C;N,u_S)=\phi ^g (C;N,u_S)\) holds for any \(C\subset N\) with \(|C|<r'\). We will check that \(\xi ^g (D;N,u_S)=\phi ^g (D;N,u_S)\) for all \(D\subset N\) with \(|D|=r'\).

Let D be a fixed subset of N of cardinality \(r'\). Since \(S\varsubsetneq N\), again there is a null player j in \(u_S\). So, if \(j\in D\), then \(D{\setminus } j\) is a coalition of cardinal \(r'-1\) and, thus, G-null player and the second induction hypothesis imply

$$\begin{aligned} \xi ^g (D;N,u_S)=\xi ^g ({D\backslash j};N,u_S)=\phi ^g ({D\backslash j};N,u_S)=\phi ^g(D;N,u_S). \end{aligned}$$

Otherwise, if D does not contain any null player, then let i be a player in \(D\subset S\). By the second induction hypothesis and G-null player property it holds

$$\begin{aligned} \xi ^g ({(D\backslash i)\cup j};N,u_S)= & {} \xi ^g ({(D\backslash i)};N,u_S)=\phi ^g ({(D\backslash i)};N,u_S)\\= & {} \phi ^g ({(D\backslash i)\cup j};N,u_S). \end{aligned}$$

Note also that \(\xi ^g (C;N\backslash i,{u_S}\vert _{N{\setminus } i}))=0=\phi ^g (C;N\backslash i,{u_S}\vert _{N{\setminus } i}))\) for all \(C\subset N\), since \(u_S\vert _{N{\setminus } i}\equiv u^0\in {\mathcal {G}}_{N{\setminus } i}\). Hence, by G-CBC, taking \(C=D\backslash i\), and the first induction,

$$\begin{aligned}&\xi ^g (D;N,u_S)- \phi ^g ({(D\backslash i)\cup j};N,u_S) \\&\quad = (\phi ^g (D;N\backslash j,{u_S}\vert _{N{\setminus } j})-\phi ^g ({D\backslash i};N\backslash j,{u_S}\vert _{N{\setminus } j})) \\&\qquad -(\phi ^g ({(D\backslash i)\cup j};N\backslash i,{u_S}\vert _{N{\setminus } i})-\phi ^g ({D\backslash i};N\backslash i,{u_S}\vert _{N{\setminus } i})) \\&\quad =\phi ^g (D;N,u_S)- \phi ^g ({(D\backslash i)\cup j};N,u_S). \end{aligned}$$

So we have proved the uniqueness for the unanimity games \((N,u_S)\) for all \(S\subset N\in {\mathcal {N}}\) and we are done. \(\square \)

Aside from the previous considerations regarding the alternative axioms of Marichal et al. (2007), it must be remarked that it is not a trivial extension of a characterization of the Shapley value, since we do not impose any condition about the value of the individual agents out of the group C we are evaluating. In particular, we have been forced to use in the same characterization group linearity and group coalitional balanced contributions properties. In the following we will check that all the axioms above are necessary to guarantee the uniqueness of the Shapley group value \(\phi ^g\).

G-null player. Let \(\alpha \in (0,1)\). Define a value \(\xi ^g\) in the following manner. If (Nv) is a null game, \(\xi ^g_C(N,v)\)=0 for every \(C\in N\). Given a non-null unanimity game \(u_S\), with \(S\subsetneq N\), define \(\xi ^g_C(u_S)\) as

$$\begin{aligned} \xi ^g_C(N,u_S)={\left\{ \begin{array}{ll} \displaystyle \sum _{k=0}^{c-1} \alpha ^{n-k}, &{} \text { if }C\cap S=\varnothing , \\ \displaystyle \frac{1}{s-\vert S\cap C\vert +1} + \sum _{k=0}^{n-s-1} \alpha ^{n-k}, &{} \text { if }C\cap S\ne \varnothing , \end{array}\right. } \end{aligned}$$
(12)

\(\xi ^g_C(N,u_N)=\frac{1}{n-c+1}\), for all \(C\subseteq N\), and then extend the value by additivity. It is clear that \(\xi ^g_i (N,u_S)=\alpha ^n >0\) for all \(i\notin S\). So, G-null player does not hold. Let us check that \(\xi ^g\) verifies G-CBC over the class of unanimity games. If \((N,u_N)\), then G-CBC condition (10) trivially holds. Let \((N,u_S)\) be a unanimity game with \(S\subsetneq N\). If \(\vert S\vert =n-1\), then two cases are possible:

  1. (a)

    If \(i,j\in S\), then \(\xi ^g_{C\cup i} (N,u_S)= \frac{1}{s-\vert S\cap C\vert } + \sum _{k=0}^{n-s-1} \alpha ^{n-k}=\xi ^g_{C\cup j} (N,u_S)\) and (10) holds, since \(u_S\vert _{N{\setminus } i}\) and \(u_S\vert _{N{\setminus } j}\) are null games.

  2. (b)

    If \(i\in S\) and \(j\notin S\), then \(S=N{\setminus } j\), \(u_S\vert _{N{\setminus } i}\equiv 0\) and \(u_S\vert _{N{\setminus } j}\) is a unanimity game w.r.t. the grand coalition \(N{\setminus } j\). Thus (10) holds:

    $$\begin{aligned}&\xi ^g_{C\cup i} (N{\setminus } j, u_{N{\setminus } j}\vert _{N{\setminus } j})-\xi ^g_{C} (N{\setminus } j, u_{N{\setminus } j}\vert _{N{\setminus } j}) \\&\quad =\frac{1}{(n-1)-(c+1)+1} -\frac{1}{(n-1)-c+1}=\xi ^g_{C\cup i} (N, u_{N{\setminus } j})-\xi ^g_{C\cup j} (N u_{N{\setminus } j}) . \end{aligned}$$

If \(\vert S\vert <n-1\), then three cases are possible:

  1. (a)

    If \(i,j\in S\), then \(\xi ^g_{C\cup i} (N,u_S)= \frac{1}{s-\vert S\cap C\vert } + \sum _{k=0}^{n-s-1} \alpha ^{n-k}=\xi ^g_{C\cup j} (N,u_S)\) and (10) holds.

  2. (b)

    If \(i\in S\) and \(j\notin S\), then

    $$\begin{aligned} \xi ^g_{C\cup i} (N{\setminus } j, u_{S}\vert _{N{\setminus } j}) = \frac{1}{s-\vert S\cap C\vert } + \sum _{k=1}^{n-s-1} \alpha ^{n-k} \end{aligned}$$

    and

    $$\begin{aligned} \xi ^g_{C} (N{\setminus } j, u_{S}\vert _{N{\setminus } j}) = {\left\{ \begin{array}{ll} \displaystyle \sum _{k=1}^{c} \alpha ^{n-k}, &{} \text { if }C\subseteq N{\setminus } S, \\ \displaystyle \frac{1}{s-\vert S\cap C\vert +1} + \sum _{k=1}^{n-s-1} \alpha ^{n-k}, &{} \text { otherwise.} \end{array}\right. } \end{aligned}$$

    Thus, if \(C\subseteq N{\setminus } S\):

    $$\begin{aligned}&\xi ^g_{C\cup i} (N{\setminus } j, u_{S}\vert _{N{\setminus } j})-\xi ^g_{C} (N{\setminus } j, u_{S}\vert _{N{\setminus } j}) \\&\quad =\frac{1}{s}+ \sum _{k=1}^{n-s-1} \alpha ^{n-k} - \sum _{k=1}^{c} \alpha ^{n-k} =\xi ^g_{C\cup i} (N, u_{S})-\xi ^g_{C\cup j} (N, u_{S}) \end{aligned}$$

    If \(C\cap S\ne \varnothing \), then

    $$\begin{aligned}&\xi ^g_{C\cup i} (N{\setminus } j, u_{S}\vert _{N{\setminus } j})-\xi ^g_{C} (N{\setminus } j, u_{S}\vert _{N{\setminus } j}) \\&\quad =\frac{1}{s-\vert S\cap C\vert }+ \sum _{k=1}^{n-s-1} \alpha ^{n-k} - \left( \frac{1}{s-\vert S\cap C\vert +1}+ \sum _{k=1}^{n-s-1} \alpha ^{n-k} \right. \\&\quad \left. =\xi ^g_{C\cup i} (N, u_{S})-\xi ^g_{C\cup j} (N, u_{S}) \right) \end{aligned}$$
  3. c)

    If \(i,j\in N{\setminus } S\), then \(C\cup i\subseteq N{\setminus } S\) if, and only if, \(C\cup j\subseteq N{\setminus } S\) and, therefore condition (10) can be easily checked.

Now, since G-CBC condition is additive \(\xi ^g\) satisfies it over \(\bigcup _{n\ge 1} {\mathcal {G}}_n\)

G-linearity. Let \(\xi ^g\) be another group value over (Nv) which is defined in the following way:

\((\mathrm{A}_1)\) :

If there is at least a null player in N, or (Nv) is the unanimity game with respect to the grand coalition N, then \(\xi ^g_C(N,v)=\phi ^g_C(N,v)\) for every group C in N.

\((\mathrm{A}_2)\) :

Otherwise, \(\xi ^g_C(N,v)=\phi ^g_C(N,v)+k\), being \(k\ne 0\) a fixed constant.

It is easily checked that G-null-player and G-SPB hold for \(\xi ^g\). We will check that the property of coalitional balanced contributions G-BMC also holds for this value. Note that all differences in (10) match \(\xi ^g_{D} (L,w)\) with \(\xi ^g_{D'} (L,w)\), for some coalition \(L\in \{N,N{\setminus } i, N{\setminus } j\}\) and some game \(w\in \{v,v_{-i},v_{-j}\}\).

Taking into account that \(\phi ^g=\xi ^g\) in case (Nv) is some of the games in the first case \((A_1)\), and in the second one the \(k's\) cancel, it holds:

$$\begin{aligned} \xi ^g_{C\cup i} (N,v)- \xi ^g_{C\cup j} (N,v)&= \phi ^g_{C\cup i} (N,v)- \phi ^g_{C\cup j} (N,v),\\ \xi ^g_{C\cup i} (N{\setminus } j,v_{-j})- \xi ^g_{C} (N{\setminus } j,v_{-j})&= \phi ^g_{C\cup i} (N{\setminus } j,v_{-j})- \phi ^g_{C} (N{\setminus } j,v_{-j}),\\ \xi ^g_{C\cup j} (N{\setminus } i,v_{-i})- \xi ^g_{C} (N{\setminus } i,v_{-i})&= \phi ^g_{C\cup j} (N{\setminus } i,v_{-i})- \phi ^g_{C} (N{\setminus } i,v_{-i}), \end{aligned}$$

G-CBC property holds for \(\phi ^g\), and so the property does so for \(\xi ^g\), and we are done.

Note that we can modify \(\xi ^g\) defining \(\xi ^g_i(N,v)=\Phi _i (N,v)\), for all \(i\in N\), and \(\xi ^g_N(N,v)=v(N)\), for all \(N\subseteq \mathbb {N}\), and the same result holds.

G-CBC. Define a value \(\xi ^g\) in the following manner. If (Nv) is a null game, \(\xi ^g_C(N,v)\)=0 for every \(C\in N\). Given a non-null unanimity game \(u_S\) with \(S\subsetneq N\), define \(\xi ^g_C(u_S)\) as

$$\begin{aligned} \xi ^g_C(u_S)={\left\{ \begin{array}{ll} 0 &{} \text { if }C\cap S=\varnothing , \\ \xi ^g_{C}(u_S)= 1, &{} \text { if }S\subseteq C,\\ |S\cap C|\times |S\backslash C| &{} \text { if }C\cap S\ne \varnothing \text { and }C\cap S\ne S, \end{array}\right. } \end{aligned}$$
(13)

\(\xi ^g_C(N,u_N)=\frac{1}{n-c+1}\), for all \(C\subseteq N\), and then extend the value by additivity.

Observe that all axioms but G-CBC hold. The unique one that needs a bit of discussion is the G-null player axiom, which holds when considering the base of unanimity games because in \(u_S\) the null players are precisely the players outside S, and therefore:

$$\begin{aligned} \xi ^g_{C\cup i} (N,u_S)&= 0= \xi ^g_{C} (N,u_S), \text { if }S\cap C=\varnothing ,\\ \xi ^g_{C\cup i} (N,u_S)&= 1= \xi ^g_{C} (N,u_S), \text { if }S\subseteq C,\text { and}\\ \xi ^g_{C\cup i} (N,u_S)&= |S\cap (C\cup i)|\times |S\backslash (C\cup i)|=|S\cap C|\times |S\backslash C|= \xi ^g_{C} (N,u_S), \text { otherwise,} \end{aligned}$$

for all player \(i\notin S\). Then, taking into account that the Harsanyi dividend \(c_S (N,v)\) of any coalition S containing null players in the game (Nv) equals zero, G-null player property holds for any n-person game \((N,v)\in {\mathcal {G}}_n\), for all \(N\subseteq \mathbb {N}\).

Let us check by means of a concrete example that the G-CBC axiom fails in this case. Consider \((N,u_S)\) with \(|N|=3\), and \(S=\{1,2\}\). In the notation of the axiom, take \(C=\{1\}\), \(i=2\), and \(j=3\). Then:

$$\begin{aligned} \xi ^g_{\{1,2\}}(N,u_S)= & {} 1, \; \xi ^g_{\{1,3\}}(N,u_S)=1\times 1=1, \;\xi ^g_{\{1,2\}}(N\backslash 3, u_S\vert _{N \backslash 3}) \\= & {} 1, \;\xi ^g_{1}(N\backslash 3, u_S\vert _{N \backslash 3})=1/2, \end{aligned}$$

and \(\xi ^g_{\{1,3\}}(N\backslash 2, u_S\vert _{N \backslash 2})=0=\xi ^g_{1}(N\backslash 2, u_S\vert _{N \backslash 2})\), because the game \((N\backslash 2, u_S\vert _{N \backslash 2})\) is null. It is clear now that the two sides of the equalities that define the axiom do not coincide in this case.

Note that we can modify \(\xi ^g\) defining \(\xi ^g_i(N,u_S)=\Phi _i (N,v)\), for all \(i\in N\), and \(\varnothing \ne S\subsetneq N\), for all \(N\subseteq \mathbb {N}\), and the same result holds. Probably it is easy to find more examples by defining the value over \(C\cap S\) (when \(C\cap S\ne \varnothing \)) as another appropriate function of \((|C\cap S|,|S\backslash C|)\).

G-symmetry over pure bargaining games. Define a value \(\xi ^g\) in the following manner. If (Nv) is a null game, \(\xi ^g_C(N,v)\)=0 for every \(C\in N\). Given a non-null unanimity game \(u_S\), define \(\xi ^g(N,u_S)\) as \(\xi ^g_C(N,u_S)=\frac{\vert C\cap S\vert }{\vert S\vert }\), for all group \(C\subseteq N\), and then extend the value by additivity.

Observe that all axioms but G-SPB hold. The G-null player axiom trivially holds when considering the base of unanimity games. Then, since \(c_S (N,v)=0\) for all S containing null players in the game (Nv), G-null player property holds in general. Moreover, G-additivity follows from the definition.

Let us check that the G-CBC axiom holds. Let (Nv) be any n-person game with \(n\ge 2\). Let C be any group in N of cardinality \(c\le n-2\), and let \(i,j\in N{\setminus } C\). Then:

  • If \(i,j\in N{\setminus } S\), then (10) holds since all the involved differences are zero because i and j are null players in the three games.

  • If \(i,j\in S\), then \(\vert (C\cup i)\cap S\vert =\vert C\cap S\vert + 1=\vert (C\cup j)\cap S\vert \). Thus, \(\xi ^g_{C\cup i} (N,u_S)- \xi ^g_{C\cup j} (N,u_S)=0\), and (10) holds because the games \((N{\setminus } i, u_S\vert _{N{\setminus } i})\) and \((N{\setminus } j, u_S\vert _{N{\setminus } j})\) are null.

  • If \(i\in S\) and \(j\notin S\), then \(\xi ^g_{C\cup i} (N,u_S)- \xi ^g_{C\cup j} (N,u_S)=\frac{1}{s}=\xi ^g_{C\cup i} (N{\setminus } j, u_S\vert _{N{\setminus } j})- \xi ^g_{C} (N{\setminus } j, u_S\vert _{N{\setminus } j})\), and (10) holds because \(u_S\vert _{N{\setminus } i}\equiv 0\).

Clearly, G-SPB fails, so we are done.

Proposition 1

Let \(N\in {\mathcal {N}}\) be any finite set of players, and v be any game in \({\mathcal {G}}_N\). Then, the Shapley group value \(\phi ^g\) verifies the following properties:

(i):

Group Rationality: \(\phi ^g (C;N,v) \ge v(C)\) for every \(C\subset N \in {\mathcal {N}}\) if the game \(v\in {\mathcal {G}}_N\) is superadditive, and

(ii):

Monotonicity: \(\phi ^g (C;N,v) \le \phi ^g (D;N,v)\) for every pair of coalitions \(C\subset D\subset N\in {\mathcal {N}}\) if the game \(v\in {\mathcal {G}}_N\) is monotonic.

Proof

Group rationality follows from the individual rationality of the Shapley value. Note that every merging game \((N_C,v_C)\), \(C\subset N\), is superadditive if it is (Nv), for all \(C\subset N\in {\mathcal {N}}\) and \(v\in {\mathcal {G}}_N\).

Monotonicity follows from being

$$\begin{aligned} \phi ^g ({C\cup i};N,v) - \phi ^g (C;N,v)&=\sum _{\begin{array}{c} S\subset N{\setminus } C\\ i\notin S \end{array}} \frac{s! (n-c-s)!}{(n-c+1)!} \Bigl ( v(S\cup i\cup C)-v(S\cup C)\Bigr ) \nonumber \\&\quad +\frac{(s+1)! (n-c-s-1)!}{(n-c+1)!} \Bigl ( v(S\cup i)-v(S) \Bigr ) \ge 0, \end{aligned}$$
(14)

for all coalitions \(C\subset N\in {\mathcal {N}}\), and all players \(i\notin C\), whenever the game v is monotonic. \(\square \)

Theorem 2

Let \(N\in {\mathcal {N}}\) be any finite set of players, and v be any game in \({\mathcal {G}}_N\). Let \(C\subset N\) be any group in N, and let \(i\notin C\). Then, the marginal contribution of player \(i\in N{\setminus } C\) to the Shapley group value of C equals:

$$\begin{aligned} MC_i^{g} (C;N,v):=\phi ^g ({C\cup i};N,v)-\phi ^g (C;N,v)=\phi _i (N{\setminus } C, v\vert _{N{\setminus } C}) + \psi _{\mathbf {c}i} (N_C,v_C). \end{aligned}$$
(15)

Proof

Let us arbitrarily fix the sets C and N and the player i, \(C\subsetneq N\in {\mathcal {N}}\), \(i\in N{\setminus } C\), and let \(v\in {\mathcal {G}}_N\). Then, adding and subtracting the amount

$$\begin{aligned} \sum _{\begin{array}{c} S\subset N{\setminus } C\\ i\notin S \end{array}} \frac{s! (n-c-s-1)!}{(n-c)!} \Bigl ( v(S\cup i)-v(S)\Bigr ) \end{aligned}$$

to the above expression (14) of the marginal contribution \(MC_i^{g} (C;N,v)\) follows that:

$$\begin{aligned} MC_i^{g} (C;N,v)=&\sum _{\begin{array}{c} S\subset N{\setminus } C\\ i\notin S \end{array}} \frac{s! (n-c-s-1)!}{(n-c)!} \Bigl ( v(S\cup i)-v(S)\Bigr ) \\&+ \sum _{\begin{array}{c} S\subset N{\setminus } C\\ i\notin S \end{array}} \frac{s! (n-c-s)!}{(n-c+1)!} \Bigl ( v(S\cup i\cup C)-v(S\cup C)\\&-v(S\cup i) + v(S)\Bigr ). \end{aligned}$$

The first term is precisely the Shapley value of player i in the restricted game \((N{\setminus } C, v\vert _{N{\setminus } C})\), where players in C do not play a role. The second term can be expressed by means of the second-order difference operators for the pair of players \(i,\mathbf {c}\in N_C\), as follows:

$$\begin{aligned}&\sum _{\begin{array}{c} S\subset N{\setminus } C\\ i\notin S \end{array}} \frac{s! (n-c-s)!}{(n-c+1)!} \Bigl ( v(S\cup i\cup C)-v(S\cup C)-v(S\cup i) + v(S)\Bigr )\\&\quad =\sum _{\begin{array}{c} S\subset N{\setminus } C\\ i\notin S \end{array}} \frac{s! (n-c-s)!}{(n-c+1)!} \Delta ^2_{ic} (S;N_C,v_C)=:\psi _{\mathbf {c}i} (N_C,v_C) . \end{aligned}$$

\(\square \)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Flores, R., Molina, E. & Tejada, J. Evaluating groups with the generalized Shapley value. 4OR-Q J Oper Res 17, 141–172 (2019). https://doi.org/10.1007/s10288-018-0380-8

Download citation

  • Received:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10288-018-0380-8

Keywords

Mathematics Subject Classification

Navigation