We use an experiment to examine escalation bias in subjective performance evaluations. Participants assume the role of manager and evaluate the performance of an employee based on a balanced-scorecard-type performance report. We manipulate whether managers recommended positively or negatively about the evaluated employee's promotion to his current position. Consistent with the presence of escalation bias, we find that managers give higher performance ratings to employees about whom they advised positively than to employees about whom they advised negatively. Using eye-tracking data, we investigate whether escalation bias arises because managers with different prior commitments toward the evaluated employee pay attention to different items in the scorecard. We find that evaluators' prior recommendation does not affect what proportion of their visual attention is given to favorable (versus unfavorable) performance measures, and that the relative attention paid to favorable measures is not associated with the performance rating.