The Estimation Accuracy of True Scores at Different Degrees of Local Dependence Among Test Items for Different Ability Distributions

Abstract

The study aimed to investigate the effect of local depdendence among items on the estimation of true scores under different frequency distribution for examinees’ ability. The study used randomly twelve data sets; each with 2000 examinees. These data sets resulted from controling two factors: ability distributions )3 distribution: normal, positively skewed, negatively skewed( and local depdence )4 levels: correlation of zero, 0.3, 0.6, 0.9(. Each data sets manipulate the local dependence for 12 items out from 60 real items )20%(. Results revealed that the estimation accuracy of true scores decreases as the local dependence increases. Generally, local dependence among items showed positive bias in the estimation of True Scores across most ability scales. Moreover, ability distribution influenced the effect of local dependence on the estimation accuracy of true scores.