Principal Component Analysis and Power Indices

Abstract

Measuring the influence of a player in a simple game is a widely studied topic. Shapley-Shubik power index is perhaps the maximum exponent in terms of relevance. Furthermore, other power indexes have been proposed over time. In this paper, we propose yet another power index, defined in terms of winning coalitions. We show that this index coincides with eigenvalues obtained with the Principal Component Analysis method, a broadly used technique in data science to determine the influence of different features given a dataset. Furthermore, we provide a characterization of this proposed index in terms of four properties.