The Effect of the Percentages of Missing Data of Anchor Test Form Including the Methods of Imputation for Lower and Upper Levels of a Test in the Accuracy of Estimated Parameters of the Ability and Items

Abstract

The research investigated the effect of the missing rate (5%, 15%, and 30%) in the data of anchor test, of lower and upper levels of a test form, including the method of imputing the missing values, on the accuracy of estimated parameters of ability and items. To achieve this, the experimental method was used. Using the 2PL model, responses of 1000 persons for both levels of the test that consisted of 80 items of 16 anchor items that were generated using WinGen-3 by 10 replications. Lava-1.6 model was used to guarantee the the loss in each item of the anchor items which are due to the adopted rates. Then, the missing values were imputed using the methods of imputation (ICC, MI, EM), where the means and standard deviations error in estimating parameters of abilities, discrimination, and difficulty at two levels had been calculated. However, the repeated measures of ANOVA came up with results showing that there were significant differences among the means of the standard error in estimating parameters of distiction, and difficulty in favor to the situations of estimated parameters of ability and difficulty to the independent variable of the research in favor of estimating parameters of discrimination (original), at α=0.05.