Arbitrariness of peer review: A Bayesian analysis of the NIPS experiment