Traditional power indices ignore preferences and strategic interaction. Equilibrium analysis of particular non-cooperative decision procedures is unsuitable for normative analysis and assumes typically unavailable information. These points drive a lingering debate about the right approach to power analysis. A unified framework that works both sides of the street is developed here. It rests on a notion of a posteriori power which formalizes players' marginal impact to outcomes in cooperative and non-cooperative games, for strategic interaction and purely random behaviour. Taking expectations with respect to preferences, actions, and procedures then defines a meaningful a priori measure. Established indices turn out to be special cases.