Examination the Effect of Missing Data Techniques of Item Parameters


Creative Commons License

SAYIN A. , YANDI A., OYAR E.

JOURNAL OF MEASUREMENT AND EVALUATION IN EDUCATION AND PSYCHOLOGY-EPOD, vol.8, no.4, pp.490-510, 2017 (Journal Indexed in ESCI) identifier

  • Publication Type: Article / Article
  • Volume: 8 Issue: 4
  • Publication Date: 2017
  • Doi Number: 10.21031/epod.355506
  • Title of Journal : JOURNAL OF MEASUREMENT AND EVALUATION IN EDUCATION AND PSYCHOLOGY-EPOD
  • Page Numbers: pp.490-510

Abstract

In this study, the aim is to determine how the item and test parameters affect the missing data techniques for different sample sizes and different items with different missing data rates. 500, 100 and 2500 students randomly selected from the 5073 students who participated in the PISA 2015 study and responded to the "ambition perception" scale included in the study constitute the study group of the research. First of all, the assumptions of normality, unidimensionality, local independence and model-data fit were examined for each data set. Afterwards, 5%, 10%, 15%, and 20% missing data were formed for four out of five items and there was no missing data in one item, then analyses were carried out. Once it is determined that the missing data are missing completely random, first with complete and incomplete data, then with serial mean, median of nearby points, mean of nearby points, linear interpolation, linear trend at point, regression, expectation maximization algorithm data item and test parameters were estimated. In the estimated process, descriptive statistics and cronbach alpha reliability coefficient and marginal reliability coefficient; the threshold parameters and the difficulty indices were estimated according to the graded response theory, which is one of the IRT models. The results of the study showed that the item and test parameters were influenced by incomplete and missing data techniques; it was determined that the best estimation results were obtained by linear interpolation method with different data.