Kayıp Veriyle Başa Çıkma Yöntemlerinin Test Eşitlemeye Etkisinin İncelenmesi

Abstract

In this study; It is aimed to examine the effects of missing data handing methods on the performance of test equating methods under different conditions. In the research; sample size (750, 1500), missing data rate (10%, 20%, 30%) and test form with missing data (both tests, test to be equalized) were taken as conditions. Performances of test equating methods; root mean square error (RMSE) and equating bias (BIAS). The full datasets used in the research were obtained from the data of students in all countries participating in TIMSS 2019 with the electronic application (eTIMSS), who took booklets 6 and 7 and answered the items scored in two categories (1-0). Zero imputation (ZI), hot deck imputation (HDI) and multiple imputation (PMM and LOGREG) methods were used to assign data to test forms with missing data, which were created by deleting data under the missing completely at random data mechanism. Completed test forms were equated with test equating methods based on IRT. All analyzes in the research were carried out using the R programming language. It was determined that the methods that produced the mean RMSE and BIAS values closest to the reference value were PMM and LOGREG, while the HDI method showed similar performance to these methods. It has been observed that the method that produces the most erroneous and biased results and makes the furthest estimations from the reference value is ZA. It was determined that the sample size condition in the research had a significant effect on the methods of handling with all missing data, while the missing data rate and the test form condition with missing data had a significant effect on the average RMSE and BIAS values obtained only by the ZI method. It has been suggested that PMM, LOGREG and HDA methods should be preferred in all conditions discussed in the study, while SA method should be preferred in large samples and low loss data rates where it performs best.