Kayıp Veriyle Başa Çıkma Yöntemlerinin Değişen Madde Fonksiyonu Üzerindeki Etkisinin İncelenmesi

Abstract

The purpose of this study is to investigate the impacts of zero imputation (ZI), multiple imputation (MI), expectation maximization (EM) on differential item functioning under different conditions. The impacts of these three methods were evaluated with regard to type I error rate and power rate. The conditions investigated in this study are sample size (450 and 900), focal-reference group rate (150/300, 225/225, 350/550, 450/450), DIF level (A, B and C) and missing data rate (%10, %20 and %30). Data sets used in this study are generated under missing completely at random (MCAR) mechanism deleting data from complete data set taken from PISA 2012 maths test. Using ZI, MI and EM methods data imputation has been done to data sets. DIF analysis has been made through SIBTEST method on data sets. In terms of type I error rates, ZI performs better at %30 missing data rate than the other rates and ZI performs better at A and B DIF levels than C DIF level; MI performs well for all the conditions which are examined and EM performs better at A DIF level condition than B and C DIF levels. It has been noticed that ZI has more acceptable power rates at % 10 missing rate condition than it has %20 and %30 missing data rates. It has been observed that the power rates of both EM and MI perform worse at %30 missing data rate condition as well as A DIF level condition than all the other conditions examined.