DIFFERENTIAL PREDICTABILITY IN THE USE OF TEST SCORES