Best Split-Half and Maximum Reliability

The paper addresses an iterative method by which a test can be dichotomized in parallel halves and ensures maximum split-half reliability. The method assumes availability of data on scores of binary items. Since, it was aiming at splitting a test in parallel halves, no assumption was made regarding form or availability of reference test. Empirical verification is also provided. Other properties of the iterative methods discussed. New measures of degree of parallelism given. Simultaneous testing of single multidimensional hypothesis of equality of mean, variance and correlation of parallel tests can also be carried out by testing equality of regression lines of test scores on scores of each of the parallel halves, ANOVA or by Mahalanobis�� 2 . The iterative method can be extended to find split-half reliability of a battery of tests. The method thus provides answer to much needed question of splitting a test uniquely in parallel halves ensuring maximum value of the split-half reliability. The method may be adopted while reporting a test.