Pearson product-moment correlation coefficient between item g and test score X, known as item-test or item-total correlation ("Rit"), and item-rest correlation ("Rir") are two of the most used classical estimators for item discrimination power (IDP). Both "Rit" and "Rir" underestimate IDP caused by the mismatch of the scales of the item and the score. Underestimation of IDP may be drastic when the difficulty level of the item is extreme. Based on a simulation, in a binary dataset, a good alternative for "Rit" and "Rir" could be the Somers' "D": it reaches the ultimate values +1 and -1, it underestimates IDP remarkably less than "Rit" and "Rir", and, being a robust statistic, it is more stable against the changes in the data structure. Somers' "D" has, however, one major disadvantage in a polytomous case: it tends to underestimate the magnitude of the association of item and score more than "Rit" does when the item scale has four categories or more.