This study aimed to investigate the effect of the percentage of missing data on estimating the standard error of the items' parameters and the test information function based on the three-parameter model using generated data. To achieve the objectives of the study, (1000) examinees' responses were generated on a test consisting of (50) dichotomous items according to the three-parameter logistic model( 3PL) using (WINGEN3) software. By using (SPSS) and (EXCEL), data containing missing responses were obtained at the percentages of (0%, 5%, 10%, 15%). Four files were prepared that include the missing four percentages. The items' parameters for these files were identified, , the items and individuals were fitted for ( 3PL-IRT), number of persons were removed.
The results of the study showed that there are statistically significant differences between the means of the standard errors for estimating the difficulty and the discrimination parameter of the test based on the three-parameter model according to the difference in the missing data percentages of (0%, 5%, 10%, 15%) and in favor of the lesser missing percentages, which means that the fewer missing percentages are, it is expected that the standard error value will decrease in estimating the difficulty and the discrimination parameter of the item, and that the best missing data percentage was (5%). The results also indicated that the information function increases as the missing data percentage decreases.
Key words: Missing Data, Standard Error, Test Information Function
|
