Towards a New Paradigm for Statistical Evidence