Consider for example a parliamentary situation in which there are n players, with the same number of votes, and a simple majority of them is required to form a government. A straightforward application of the Shapley value grants each player 1/n, using symmetry considerations. In real-life parliaments, however, it is intuitively clear to all observers that not all members have equal power. It is highly unusual to see, for example an extreme right party joining an extreme left party in a coalition without any center parties also included in the coalition to bridge political differences between them.
As the above discussion indicates, part of the problem is that the standard Shapley approach assumes that all possible permutations of the players be used in forming coalitions. That means that even highly unlikely coalitions, such as those formed by an extreme left party joining with an extreme right party while by-passing all the parties in between, including their most natural political allies, must be counted equally along with every other coalition.
Different approaches have been proposed in the literature to study situation in which not all coalitions are feasible or equally likely. In many papers, the problem is tackled by considering some structure on the set of players to describe the way in which players can form coalitions. Coalitional games together with these kind of structures are usually denoted games with restricted cooperation.
One of the most widely-studied model of games with restricted cooperation is the restricted communication model proposed by Myerson (1977). In Myerson’s approach, in addition to the game itself one considers an undirected graph that describes communication possibilities between the players. A modification of the Shapley value is then proposed under the assumption that coalitions that are not connected in this graph are split into connected components. In contrast, in our approach we focus on permutations, that is, on the way in which coalitions are formed, instead of imposing restrictions directly on possible coalitions.
We propose here an intuitive way to modify the Shapley value by taking the political spectrum explicitly into account. The incorporation of the ideological positions of the agents for the study of the power distribution of a decision-making body was first introduced by Owen (1972). In that work, agents’ political positions are given as points in a high-dimensional Euclidean space, and a probability distribution on the set of all permutations is inferred from them. Then, a modification of the Shapley value is proposed based on two properties, namely that an ordering and its reverse ordering should have the same probability and that the removal of a subset of agents should not affect the probabilities assigned to the relative orderings of the remaining agents.
Shapley (1977) proposed taking into account the political positions of the agents as well, using this to develop an asymmetric generalization of the Shapley value. This modification of the original Shapley value was also considered in Owen and Shapley (1989) to study the optimal ideological position of candidates. More recently, Alonso-Meijide et al. (2011) introduced what they termed the distance index. This value for simple games is another modification of the Shapley value that takes into account the ideological positions of agents. Based on Euclidean distances between agents, a probability distribution is constructed that gives high probability to coalitions whose total distance is relatively small.
Even though it is based on ideas similar to the above-cited works, our approach is much simpler. Firstly, we consider only ordinal positions in a one-dimensional space, without further exogenous specification of distances. Secondly, we assign equal probability to all the permutations that are admissible in our model. This simplicity allows for a characterization of the value by means of a set of properties and eases computation of the value. With regard to the properties of the Owen (1972) value, the value introduced here shares the first of those properties but not the second one.
In this work we assume that there exists a spectrum, from ‘left to right’ according to which the players are ordered linearly. We then impose the condition that as coalitions are formed à la Shapley, they must be connected with respect to the spectrum. Hence, we propose a novel way to generalize the Shapley value to games with restricted cooperation in which the restrictions arise from the position of the agents in a one-dimensional spectrum. This leads to an interesting new value that may shed light on relative power measures in situations in which there is a natural ordering of the players.
Perhaps the paper with the most similar general motivation to ours is Gilles and Owen (1999), in which an exogenously given hierarchy amongst players is assumed (as opposed to the exogenously given spectrum as in our paper). A player in