Abstract

Popper's idea of propensities constituting the physical background of predictable probabilities is reviewed and developed by introducing a suitable formalism compatible with standard probability calculus and with its frequency interpretation. Quantum statistical ensembles described as pure cases (“eigenstates”) are shown to be necessarily not homogeneous if propensities are actually at work in nature. An extension of the theory to EPR experiments with local propensities leads to a new and more general proof of Bell's theorem. No joint probabilities for incompatible observables need to be introduced.