Forecasting the Bayes factor of a future observation

I present a new procedure to forecast the Bayes factor of a future observation by computing the predictive posterior odds distribution. This can assess the power of future experiments to answer model selection questions and the probability of the outcome, and can be helpful in the context of experiment design. As an illustration, I consider a central quantity for our understanding of the cosmological concordance model, namely, the scalar spectral index of primordial perturbations, n(s). I show that the Planck satellite has over 90 per cent probability of gathering strong evidence against n(s) = 1, thus conclusively disproving a scale-invariant spectrum. This result is robust with respect to a wide range of choices for the prior on ns.