Generalizing from ‘Imperfect’ Replication