Methods to convert test scores into integer grades were discussed from both the classical and the IRT points of view. Using the classical test theory, a method to convert test score into the integer grade such that the resulting confidence interval of the true score has a prescribed width, say, 1 or 2, was presented, and a method to define the optimal number of grades was derived. From IRT point of view, it was shown that, if we calculate the average standard error of measurement, the similar method can be used to define the optimal number of grades. Also, a method to utilize the variance-stabilizing transformation was proposed.
View full abstract