Kayıp Veri ile Baş Etme Yöntemlerinin Güvenirlik Kestirimleri Üzerine Etkisi

Abstract

The purpose of this study was to compare the missing data handling methods in terms of the influence on reliability coefficients under different simulation conditions. For this purpose, datasets with 200, 500, and 1000 sample sizes, 20 and 40 items were generated, and study was conducted with those simulated datasets. Cronbach alpha and IRT based reliability coefficients of each dataset was calculated and used as reference values. Afterwards, datasets were reduced by %5, %10 and %20 as completely random and generated new datasets. Using listwise deletion (LD), zero imputation, regression imputation (RI), multiple imputation (MI), and expectation maximization (EM) missing data handling methods were applied to all datasets. Cronbach alpha and IRT based empirical reliability coefficients of each dataset were calculated and compared with the reference values. As a result of the study, there was no statistically significant difference between reliability coefficients. The methods which produced the closest values to the reference values were RI and MI methods.