When temporal contiguity matters: A moderator of the predictive validity of implicit measures