Peer review versus bibliometrics: Which method better predicts the scholarly impact of publications?

In this work, we try to answer the question of which method, peer review versus bibliometrics, better predicts the future overall scholarly impact of scientific publications. We measure the agreement between peer review evaluations of Web of Science indexed publications submitted to the first Italian research assessment exercise and long-term citations of the same publications. We do the same for an early citation-based indicator. We find that the latter shows stronger predictive power, i.e. it more reliably predicts late citations in all the disciplinary areas examined, and for any citation time window starting 1 year after publication.