Studies in item analysis. 2: Effects of various methods upon test reliability

Abstract

Four different methods of item analysis were compared, using test reliability as the criterion for evaluation. The methods were: "D" values, Davis' r , Phi-coefficient, and Per Cent Method. 150 multiple-choice vocabulary items were used. It was found that the Per Cent Method had the lowest reliability coefficients 75% of the time, and the highest reliability coefficients 12.5% of the time, while the other 3 methods were practically equal in the number of times they appear in each rank. It was also found that increasing test length raises test reliability so long as good items are being added, that most of the items selected by any one method will be selected by the others, and that size of criterion group does not affect test reliability. (PsycInfo Database Record (c) 2025 APA, all rights reserved)