The Definition and Estimation of Test Reliability

empirical judgment, and different authors have given conflicting definitions and conflicting criteria to guide empirical judgment. Under several of these definitions, however, it can be shown that the reliability coefficient is the variance ratio of true scores to total scores on each of the equivalent forms. The present writer prefers to reverse this procedure, defining the reliability coefficient of any one form of a test as the variance ratio of true scores to total scores. This shifts the ground of a difficulty indicated by Kelly [6], that under the correlation definition we cannot discuss directly the reliability of one form of a test. Under the proposed definition such discussion is entirely possible so long as it remains formal, and we see that the second form is needed only when we come to the estimation of the