Investigating the Effect of Test Length and Sample Size on the Accuracy Methods for Estimating Items Parameters and Persons Abilities in the Bilog Program

Abstract

This study aimed at investigating the effect of test length and sample size on the accuracy methods for estimating item parameters and persons abilities in the program Bilog. To achieve this goals 100 replicated binary responses through three- parameter model were generated using WINGEN program on four test lengths )10, 30 ,60, 80( and four sample sizes )250, 500, 1000, 1500(. And using estimation methods )MML, EAP, MAP( used in Bilog program the generated data were analyzed by Bilog program to estimate the item parameter(discrimination, difficulty, guessing), the abilities of persons, and the computed values of RMSE. The results revealed significant differences due to sample size and test length and their interaction in the accuracy estimates of item parameter when using marginal maximum likelihood )MML(, and the accuracy of item parameter increased when increasing test length and sample size. The results also revealed that when calibrating the person ability using three estimation methods used in Bilog program there is significant differences due to sample size, test length and estimation method and their interaction in the accuracy estimates of person ability, also the method )EAP( is better accurate than other estimation methods of person ability. In general the results revealed the accuracy item parameters and abilities of persons increased as the test length exceeded 30 items and the sample size exceeded 500 subjects.