Assessing evidence for replication: A likelihood-based approach