Abstract
This paper introduces a class of cooperative allocation rules for TU-games with a network structure that account for an agent’s centrality in the network. In contrast to existing centrality measures, this approach analyzes the consequences of tie failure (rather than node failure). Though not directly applied to the centrality issue, tie failures are the idea underlying the Position value (the Shapley value of the arc-game). In contrast, we allow for a whole class of allocation rules as a basis since the Shapley approach might be unreasonable in political applications, especially in networks with incompatibilities. Requiring additivity (A), equal treatment of equals (Symmetry S), and irrelevance of unproductive agents (nullplayer irrelevance N) for the underlying basis, we define the class of ASN-position values and provide axiomatic characterizations. To emphasize the crucial role of links, we refine this class to multiplicative ASN-position values and provide a monotonicity result with respect to links. We apply our approach to the case of the 2001 state parliament elections in Hamburg, Germany to use our allocation rule as a power index and further provide an example where we use our approach as a centrality measure to identify top key nodes. In both applications, the position value taking the Banzhaf value as a basis turns out to be more convincing than the Shapley-based approach.
Similar content being viewed by others
Notes
In the setting of cooperative games on networks, links from an individual to itself, that is, links ii are excluded. Therefore, we add \(i\ne j\) to the definition.
Imposing for example Efficiency would uniquely determine the Shapley value. Then, single-valuedness is obsolete and NI could be replaced by either N or NPO.
This axiom is also know as arc anonymity (e.g. van den Nouweland 1993), or link anonymity (e.g. van den Nouweland and Slikker 2012) which might be more convincing terms if no relation to centrality is of interest. We stick to the original term due to Borm et al. (1992) in order to emphasize the centrality issue.
Note that such direct characterizations for general networks may be possible within the framework of Jackson and Wolinsky (1996) if one further demands superadditive coalition functions as in van den Nouweland and Slikker (2012) or Belau (2016). However, the Jackson and Wolinsky (1996) framework explicitly differs from the framework of communication situations we use here such that findings cannot be transferred. Furthermore, superadditivity is quite a restrictive property, especially for political applications and in presence of incompatibilities.
Note that Slikker (2005b) uses reward games which differ from TU-games with a network structure, such that we need to slightly modify the definition.
An axiomatization of the multiplicative normalized Banzhaf value is, for example, proposed by van den Brink and van der Laan (1998).
For completeness, one should note that this coalition broke 2 years later due to personal issues between the leaders of “CDU” and “Schill.” However, these issues were absent at the time of coalition formation, and hence, cannot be taken into account for forecasting issues.
For further details see the appendix, Table 8.
References
Banzhaf J (1952) Weighted voting doesn’t work: a mathematical analysis. Rutgers Law Rev 19(2):317–343
Barrat A, Barthélémy M, Pastor-Satorras R, Vespignani A (2004) The architecture of complex weighted networks. Proc Natl Acad Sci 101(11):3747–3752
Belau J (2013) A new outside option value for networks: the kappa-value. Econ Theory Bull 1(2):175–188
Belau J (2016) Outside option values for network games. Math Soc Sci 84:76–86
Bergantiños G, Carreras F, García-Jurado I (1993) Cooperation when some players are incompatible. Zeitschrift für Oper Res 38(2):187–201
Bonacich P (1972) Factoring and weighting approaches to status scores and clique identification. J Math Sociol 2(1):113–120
Borm P, Owen G, Tijs S (1992) On the position value for communication situations. SIAM J Discret Math 5:305–320
Coleman JS (1971) Control of collectivities and the power of a collectivity to act. In: Lieberman B (ed) Social Choice. Gordon and Breach, New York, pp 269–300
Derks JM, Haller H (1999) Null players out? Linear values for games with variable supports. Int Game Theory Rev 1:301–314
Davis M, Maschler M (1965) The kernel of a cooperative game. Naval Res Logist Q 12(3):223–259
Dubey P, Shapley L (1979) Mathematical properties of the banzhaf power index. Math Oper Res 4:99–131
Edgeworth FY (1881) Mathematical psychics: an essay on the application of mathematics to the moral sciences. Paul, C. K., London
Freeman LC (1978) Centrality in social networks: conceptual clarification. Soc Netw 1:215–239
Ghintran A (2013) Weighted position values. Math Soc Sci 65:157–163
Ghintran A, González-Arangüena E, Manuel C (2012) A probabilistic position value. Ann Oper Res 201(1):183–196
Gillies DB (1959) Solutions to general non-zero-sum games. In: Tukcer AW, Luce RD (eds) Contributions to the theory of games, vol 4. Princeton University Press, Princeton
Gómez D, González-Arangüena E, Manuel C, Owen G, Del Pozo M, Tejada J (2003) Centrality and power in social networks: a game theoretic approach. Math Soc Sci 146(1):27–54
Grofman B, Owen G (1982) A game theoretic approach to measuring degree of centrality in social networks. Soc Netw 4(3):213–224
Haeringer G (2006) A new weight scheme for the shapley value. Math Soc Sci 52(1):88–98
Jackson M, Wolinsky A (1996) A strategic model of social and economic networks. J Econ Theory 71:44–74
Kamijo Y (2009) A linear proportional effort allocation rule. Math Soc Sci 58:341–353
Kamijo Y, Kongo T (2009) Weighted position value. GLOPE II working paper series no. 16. Waseda University, Tokyo
Meessen R (1988) Communication games. Master’s thesis, Department of Mathematics, University of Nijmegen, The Netherlands
Myerson R (1977) Graphs and cooperations in games. Math Oper Res 2:225–229
Newman MEJ (2004) Analysis of weighted networks. Phys Rev E 70:056131
Opsahl T, Colizza V, Panzarasa P, Ramasco JJ (2008) Prominence and control: the weighted rich-club effect. Phys Rev Lett 10:168702
Owen G (1975) Multilinear extensions and the banzhaf value. Naval Res Logist Q 22:741–750
Owen G (1986) Values of graph restricted games. SIAM J Algebraic Discret Methods 7:210–220
Penrose L (1946) The elementary statistics of majority voting. J R Stat Soc 109(1):53–57
Ruiz LM, Valenciano F, Zarzuelo JM (1998) The family of least square values for transferable utility games. Games Econ Behav 24(1):109–130
Shapley L (1953) A value for n-person games. In: Kuhn H, Tucker A (eds) Contributions to the theory of games, vol II. Princeton University Press, Princeton , pp 307–317
Shapley L, Shubik M (1969) Pure competition, coalitional power and fair devision. Int Econ Rev 10(3):337–362
Slikker M (2005a) A characterization of the position value. Int J Game Theory 33(4):504–514
Slikker M (2005b) Link monotonic allocation schemes. Int Game Theory Rev 7(4):473–489
Suri N, Narahari Y (2008) Determining the top-k nodes in social networks using the shapley value. AAMAS 2008: proceedings of the seventh international joint conference on autonomous agents and multi-agent systems, pp 1509–1512
van den Brink R, van der Laan G (1998) Axiomatizations of the normalized banzhaf value and the shapley value. Soc Choice Welf 15(4):567–582
van den Nouweland A (1993) Games and graphs in economic situations. Ph.d. thesis, Tilburg University, Tilburg, The Netherlands
van den Nouweland A, Slikker M (2012) An axiomatic characterization of the position value for network situations. Math Soc Sci 64(3):266–271
Wasserman S, Faust K (1994) Social network analysis: methods and applications. Cambridge University Press, Cambridge
Acknowledgements
Thanks to André Casajus, Sebastian Garmann, Linda Hirt-Schierbaum, Eliane Lambertz, Walter Trockel and Wolfgang Leininger for helpful comments and discussions, and to the participants of the Dortmund Brown Bag Seminar in 2014, the 7th RGS Doctoral Conference in Economics 2014 in Dortmund, the 10th Spain-Italy-Netherlands Meeting on Game Theory (SING10) 2014 in Krakow and the 29th Annual Congress of the European Economic Association 2014 in Toulouse. Further thanks to two anonymous reviewers and the associate editor for their valuable comments and suggestions. All remaining errors are mine.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Proofs
Proof of Lemma 1: If \(Y\in \mathcal {Y}\) satisfies S, the corresponding Y-Position value \(\pi ^Y\) satisfies DEG.
Proof
Consider any (N, v, g) which is link anonymous, that is, there exists a function \(f:\{0,1,\ldots ,|g|\}\longrightarrow {\varvec{\mathbb {R}}}\) such that \(v^N(g')=f(|g'|)\) for all \(g'\subseteq g\).Then, for any two links \(\lambda ,\tilde{\lambda }\in g\), we have
that is, all \(\lambda ,\tilde{\lambda }\in g\) are (pairwise) symmetric in \((g,v^N)\). Therefore, for any allocation rule \(Y\in \mathcal {Y}\) which satisfies S, we have \(Y_{\lambda }(g,v^N)=Y_{\tilde{\lambda }}(g,v^N)=:\alpha \in {\varvec{\mathbb {R}}}\) (constant due to S). Hence, for the corresponding Y-Position value we have for any \(i\in N\)
which proves DEG. \(\square \)
Proof of Lemma 2: If \(Y\in \mathcal {Y}\) satisfies NI, the corresponding Y-Position value \(\pi ^Y\) satisfies SLP.
Proof
Consider any (N, v, g). For all \(\tilde{\lambda }\in g\) which are superfluous in (N, v, g), that is, \(v^N(g'\cup \tilde{\lambda })=v^N(g')\) for all \(g'\subseteq g{\setminus }\{\tilde{\lambda }\}\), we have that \(\tilde{\lambda }\) is a Nullplayer in \((g,v^N)\). By Y satisfying NI=NPO+N we have for all \(i\in N\)
which proves SLP. \(\square \)
Proof of Theorem 1: If an allocation rule for TU games with a network structure satisfies A, DEG, SLP and Y-CLP, \(Y\in ASN\), it is uniquely determined on cycle-free networks and it is given by the ASN-Position value \(\pi ^Y\in \Pi (ASN)\) corresponding to Y.
Proof
Existence: Since \(\pi ^Y\in \Pi (ASN)\), that is, the corresponding Y satisfies A, S, NI, we know that \(\pi ^Y\) satisfies A, DEG and SLP by Corollary 1, Lemmas 1 and 2, respective ly. Y-CLP follows by Corollary 2.
Uniqueness: We follow the idea of the proof for the (Shapley-) Position value of Borm et al. (1992). Let Y in ASN and let \(W\in \mathcal {Y}_G \) satisfy A, DEG, SLP and Y-CLP. Consider any TU-game with a network structure (N, v, g) such that g is cycle free. The unanimity games \(\{u_T\}_{T\in 2^N{\setminus } \{\emptyset \}}\) with \(u_T(K)=1\), if \(T\subseteq K\) and 0, otherwise, build a basis of V. Hence, every \(v\in V\) can be written as
where the coefficients \(\{\mu _T\}_{T\in 2^N{\setminus } \{\emptyset \}}\) are called the Harsanyi dividends of v. Therefore, by A, it is sufficient to show that \( W(N,\beta u_T,g) \) is uniquely determined for all \(\beta \in {\varvec{\mathbb {R}}}\) and \(T\in 2^N\) such that \(|T|\ge 2\). Let such \(\beta , T\) be arbitrary but fixed.
Case 1: \(\not \exists \) \(C\in \mathcal {C}(N,g)\) such that \(T\subseteq C\).
That is, there exists \(i,j\in T\) being unconnected in g and hence, \(\beta u_T^N(g')=0\) for all \(g'\subseteq g\). Therefore, every \(\lambda \in g\) is superfluous, and hence, by SLP, we have \(W(N,\beta u_T,g)=W(N,\beta u_T,g{\setminus }\lambda _1)=W(N,\beta u_T,g{\setminus }\{\lambda _1,\lambda _2\})=\cdots =W(N,\beta u_T,\emptyset )\).
Trivially, the game \((N,\beta u_t,\emptyset )\) is link anonymous and hence, by DEG, there exists a constant \(\alpha \in {\varvec{\mathbb {R}}}\) such that
Case 2: \(\exists \) \(C\in \mathcal {C}(N,g)\) such that \(T\subseteq C\).
Consider the (unique) connected hull (cf. Owen 1986) of T, H(T), given by
As g is cycle-free, H(T) is the minimal set of nodes that are essential to connect T. Note that cycle-freeness is essential here: if there is more than one path connecting T, the intersection is empty on the disjoint parts of the connecting paths. We have
All links \(\lambda \notin g|_{H(T)}\) are superfluous in \((N,\beta u_T,g)\), hence, by SLP,\(W(N,\beta u_T,g)=W(N,\beta u_T,g|_{H(T)})\).
The game \((N,\beta u_T,g|_{H(T)})\) is link anonymous (all links have the same number of swings, namely, one) with
hence, by DEG, there exists a constant \(\alpha \in {\varvec{\mathbb {R}}}\) such that
It directly follows that
By (2) and Y-CLP, we have
Note that all \(\lambda \in g{\setminus } g|_{H(T)}=g|_{N{\setminus } H(T)}\) are Nullplayers in \((g,\beta u_T^N)\) which yields and \(Y_{\lambda }(g,\beta u_T^N)=0\) since Y satisfies N. Furthermore, all \(\lambda ,\tilde{\lambda }\in g|_{H(T)}\) are pairwise symmetric in \((g,\beta u_T^N)\) and since Y satisfies S, there exist \(B^Y_T\in {\varvec{\mathbb {R}}}\) such that
where \(B^Y_T\in {\varvec{\mathbb {R}}}\) is uniquely determined by \(Y, \beta \) and T due to single-valuedness of Y and independent on g by NPO.
On the other hand, by (1), we have
Combining this, we get
and hence
which is uniquely determined, given \(Y\in ASN \). \(\square \)
Proof of Lemma 3: If \(Y\in ASN \), the corresponding Y-Position value \(\pi ^Y\) satisfies BLC.
Proof
Let (N, v, g) be a TU game with a network structure. Following Slikker (2005a), there exists a unique linear combination of unanimity games (on link sets) representing the link game \(v^N\), that is \(v^N\) exists \(\{\beta ^{v}_{g'}\}_{g'\subseteq g}\) such that
Note that the dividends \(\beta ^v\) do not depend on the choice set g (cf. Slikker 2005a) and that \(\beta ^v_{\emptyset }=0\) for zero-normalized v.
Now let \(Y\in ASN\) and consider \((g,\beta ^v_{g'}u_{g'})\) with \(g'\subseteq g\) arbitrary but fixed. All \(\lambda \in g{\setminus } g'\) are Nullplayers in this game which yields \(Y_\lambda (g,\beta ^v_{g'}u_{g'})=0\) for all \(\lambda \in g{\setminus } g'\) by N and \(Y_\lambda (g,\beta ^v_{g'}u_{g'})=Y_\lambda (g',\beta ^v_{g'}u_{g'})\) for all \(\lambda \in g'\) by NPO. All \(\lambda ,\tilde{\lambda }\in g'\) are (pairwise) symmetric which yields \(Y_\lambda (g,\beta ^v_{g'}u_{g'})=Y_\lambda (g',\beta ^v_{g'}u_{g'})=B_{g'}^{v,Y}\) by S where \(B_{g'}^{v,Y}\in \mathbb {R}\) is unique by single-valuedness and does not depend on the choice set g (NPO).
By A we obtain
For the corresponding Y-Position value we hence have
and therefore (using that \(B_{g'}^{v,Y}\) does not depend on g) we obtain
which proves BLC. \(\square \)
Proof of Theorem 2: If an allocation rule for TU games with a network structure satisfies BLC and Y-CLP, \(Y\in ASN \), it is uniquely determined for general networks and it is given by the ASN-Position value \(\pi ^Y\in \Pi (ASN)\) corresponding to Y.
Proof
Existence: Since \(\pi ^Y\in \Pi (ASN)\), that is, the corresponding Y is an ASN-value, we know that \(\pi ^Y\) satisfies BLC by Lemma 3. Y-CLP follows by Corollary 2. Uniqueness: Suppose V, W being two allocation rules for network structures satisfying BLC and Y-CLP, \(Y\in ASN \). We proceed by induction over |g|.
Induction basis [IB]: For \(|g|=0\), that is, \(g=\emptyset \), we have that \(\mathcal {C}(N,g)=\{i\}_{i\in N}\) and hence, by Y-CLP, we have
Now suppose that \(V(N,v,g)=W(N,v,g)\) for all g such that \(|g|=k, k\ge 0\) (induction hypothesis [IH]).
Consider g such that \(|g|=k+1\). As \(k+1\ge 1\), there exists \(i,j\in N, i\ne j\) such that \(j\in C_i(N,g)\) (for all l with \(|C_l(N,g)|=1\), we have \(V_l(N,v,g)=W_l(N,v,g)\) by Y-CLP). By BLC, we have for all \(i,j\in C_i(N,g)\):
and hence
Summing up (3) over all \(j\in C\) yields:
which finishes the proof. \(\square \)
Proof of Theorem 3: Any multiplicative Y-Position value \(\pi ^Y\in \Pi (ASN_m)\) can be written as
where the weighting scheme \(\left\{ w^Y_{|g|,|g'|}\right\} _{|g'|=0,\ldots ,|g|-1}\) satisfies
-
1.
\(w^Y_{|g|,|g'|}=Y_{\lambda }(g,1_{g'\cup \lambda })\) for any \(\lambda \in g\) and \(g'\subseteq g{\setminus }\lambda \)
-
2.
\(w^Y_{|g|,|g'|}+w^Y_{|g|,|g'|+1}=w^Y_{|g|-1,|g'|}\quad \forall \,|g'|=0,\ldots ,|g|-1\)
Proof
Let \(Y\in ASN, K\subseteq N\) be arbitrary but fixed and consider the game \((N,1_K)\). All players in K as well as all players in \(N{\setminus } K\) are (pairwise) symmetric in this game, hence, by S, there must exist \(w^Y(N,K)\) and \(\tilde{w}^Y(N,K)\) (uniquely determined by single-valuedness) such that we have
for all \(i\in N\). Let \(i\in N\) be arbitrary but fixed and, for all \(K\subseteq N{\setminus }\{i\}\), consider the game \((N,1_{K\cup \{i\}}+1_K)\). One can easily check that i is a Nullplayer in this game and using N and A we observe
Now, let \(i,j\in N, j\ne i\) be arbitrary but fixed and, for all \(K\subseteq N{\setminus }\{i,j\}\), consider the game \((N,1_{K\cup \{i\}}+1_{K\cup \{j\}})\). One can easily check that i and j are symmetric in this game and using S, A and (5) we observe
For any \(K,K'\subseteq N\) with \(|K|=|K'|\), we can build a sequence\(\{K=K_0,K_1,\ldots , K_{l}=K'\}\) where every set can be obtained from its predecessor by replacing exactly one player. Using (6) we hence obtain that the weights only depend on sizes of coalitions (cf. also Ruiz et al. (1998), proof of Lemma 9). We obtain
Now let \(Y\in ASN_m\) and (N, v) be any TU-game. Note that any \(v\in V\) can be written as \( v=\sum _{K\subseteq N}v(K)\cdot 1_K. \) Using (7) and that Y is not only additive but also multiplicative (m), we have
Now let (N, v, g) be any TU-game with a network structure. Using (8), we obtain
where \(w^Y_{|g|,|g'|}=Y_{\lambda }(g,1_{g'\cup \lambda })\) for any \(\lambda \in g\), and \(g'\subseteq g{\setminus }\lambda \).
To show the second part of the Theorem, we will make use of NPO. Let \(\lambda ,\lambda '\in g\) with \(\lambda '\ne \lambda \) being a Nullplayer in \((g,v^N)\). Following the same steps as in Derks and Haller (1999) we obtain
\(\square \)
Proof of Theorem 4: We will make use of the following expression:
Lemma 6
The weights from Theorem 3 satisfy
The proof follows by induction over n (starting at \(n=m\)).
Proof of Theorem 4.1: Any multiplicative ASN-Position value \(\pi ^Y\in \Pi (ASN_m)\) satisfies LMON for all TU-games with a network structure such that the corresponding link-game is link-convex.
Proof
Let \(Y\in ASN_m\) and let (N, v, g) be any TU-game with a network structure such that the corresponding link-game \(v^N\) is link-convex. By Theorem 3, for any \(i\in N\) we have
Now consider any \(\tilde{g}\subseteq g\). We can rewrite any \(g'\subseteq g\) as \(g'=h_1\cup h_2\) with \(h_1\subseteq \tilde{g}\) and \(h_2\subseteq g{\setminus }\tilde{g}\). Note that we either have \(h_1\ne \emptyset \) or \(h_2\ne \emptyset \). From (\(*\)) and using link convexity of \(v^N\) we obtain
\(\square \)
Proof of Theorem 4.2: Any multiplicative ASN-Position value \(\pi ^Y\in \Pi (ASN_m)\) satisfies Component Decomposability CD:
Proof
Since \(g_i\subseteq (g|_{C_i(N,g)})_i\), it is sufficient to show that any ASN value Y satisfies
Let \(Y\in ASN_m, (N,v)\) be a standard TU-game and let g be a network. Consider \(C\in \mathcal {C}(N,g)\). For any \(\lambda \in g|_C\), marginal contributions are not effected by connections outside C due to the form of \(v^N\). Using this and following the same decomposition method as in the proof of Theorem 4.1, we obtain
which finishes the proof. \(\square \)
Proof of Theorem 4.3: Any multiplicative ASN-Position value \(\pi ^Y\in \Pi (ASN_m)\) can be written as
where \(\{\beta _{g'}^v\}_{g'\subseteq g}\) are the Harsanyi dividends representing \(v^N\) and \(w^Y\) is the weight defined in Theorem 3.
Proof
Let \(Y\in ASN_m\) and (N, v, g) be any TU game with a network structure. Recall from the proof of Theorem 3, that the link game \(v^N\) can be written as a (unique) linear combination of (link set) unanimity games where the corresponding scalars \(\{\beta ^v_{g'}\}_{g'\subseteq g}\) are called Harsanyi dividends. Let \(g'\subseteq g\) be arbitrary but fixed. By N, we have \(Y_{\lambda }(g,u_{g'})=0\) for all \(\lambda \notin g'\). Consider any \(\lambda \in g'\). Note that any (link set) unanimity game can be written as
Using the weights from Theorem 3 and by Lemma 6, we obtain
which yields
\(\square \)
Proof of Lemma 5: The two characterizing axioms of the ASN-Position values are independent.
Proof
To show independence of the axioms, we need to find elements in \(\mathcal {Y}_G\) which satisfy one but not the other property. The Shapley-Position value satisfies BLC but (trivially) does not satisfy CLP w. r. t. the Banzhaf value. Now consider the allocation rule which shares the Banzhaf-link power of a component equally within this component, that is, the rule given by
By definition and using Theorem 4.2, W satisfies CLP w. r. t. the Banzhaf value. To see that W does not satisfy BLC, consider the following TU-game with a network structure:
All players are connected in (N, g) resulting in \(C_i=N\) for all \(i\in N\). For the Banzhaf values of links we obtain \(Ba_{\{12\}}=\frac{3}{2^{3-1}}=\frac{3}{4}\) (3 possibilities of creating 1 when materializing) and \(Ba_{\{23\}}=Ba_{\{34\}}=\frac{1}{4}\) (1 possibility of creating 1 when materializing) resulting in a component power of . Equal sharing implies for \(i=1,2,3,4\). For BLC consider players 1 and 2. We have
In the restricted game \((\{2,3,4\},v|_{\{2,3,4\}}, \{23,34\})\), each link has a Banzhaf power of \(\frac{1}{2^{2-1}}\) and equal sharing implies for \(i=2,3,4\). In the restricted game \((\{1,2\},v|_{\{1,2\}},\{12\})\), the link has a Banzhaf power of \(\frac{1}{2^{1-1}}\) and equal sharing implies for \(i=1,2\). We hence obtain \(-\frac{1}{48}\) in the first case but \(\frac{1}{8}\ne -\frac{1}{48}\) in the second case. \(\Rightarrow \) BLC is violated. \(\square \)
Appendix B: Supplementary material political example
Rights and permissions
About this article
Cite this article
Belau, J. The class of ASN-position values. Soc Choice Welf 50, 65–99 (2018). https://doi.org/10.1007/s00355-017-1074-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00355-017-1074-4