ABSTRACT
As previously indicated, the original approach to the problem of reliability by Spearman was based on the correlation of parallel tests. Kelley (1942) has pointed out that, according to this concept, the major function of the reliability coefficient is to evaluate the judgment of the test constructor, to indicate whether or not two forms thought to measure the same thing do in fact measure approximately the same thing. Recently there have been several other approaches designed to measure the homogeneity of the items in a test. It should be noted that, if two tests have each a high “homogeneity index” while the correlation between them is low, we have a distinctly disturbing situation. The indication would perhaps be that a homogeneous field existed but that the test constructor did not know enough about that field to construct two parallel tests, clearly an unsatisfactory Situation. Likewise, suppose the “homogeneity index” is very low, but the test constructor is able to set up a different form, a parallel test that correlates highly with the first form. Here it would seem that the Situation is satisfactory. The field is not unitary, but the test constructor knows the field well enough to set up different tests and have them agree. In short, if a parallel form reliability is high, the Situation is satisfactory; if the parallel form reliability is low, the Situation is unsatisfactory, regardless of what happens to the index of homogeneity.
