No label defined (Q4263576)

The paper is concerned with interaction indices, that can be applied either to players in a cooperative game or to criteria in a decision making problem. More precisely, the framework suggested by the authors is a recently introduced family of cardinal-probabilistic interaction indices. Such an index is defined in terms of (formal) derivatives and weighting coefficients, and provides a measure of the interaction between the members of any coalition in a given game. As particular cases, interaction indices a la Shapley and a la Banzhaf arise that reduce to the respective classical value when applied to singletons. A third index, the chaining interaction index \(I_R^\nu\), is defined by considering maximal chains in the lattice \(2^N\), where \(N\) is the set of agents or criteria. It is shown that this new index is a member of the cardinal-probabilistic family by explicitly giving its weighting coefficients. For any game \(\nu\), \(I_R^\nu\) becomes a representation of \(\nu\), in the sense that they define each other. The proof uses the fact that the Harsanyi dividend function is also a representation of the game, but explicit conversion formulas between \(I_R^\nu\) and \(\nu\) are also provided and, even, between \(I_R^\nu\) and the Shapley and Banzhaf interaction indices. Finally, \(I_R^\nu\) is clearly related to the multilinear extension of game \(\nu\) and to the potential functions of the Shapley and Banzhaf values. For the entire collection see [Zbl 0912.00033].