The effect of imperfect correlation in the relationship between two sets of measurements, i.e., the tendency for predicted scores to lie closer to the mean than do the scores used to predict them -- also, the statistical technique used when one or more measures are used to predict or make a least squares estimate of scores on another measure