The Influence Of Different Missing Data Techniques On The Accuracy Of Parameter Estimation For The Two-Parameter Logistic Model

The Item Response Theory makes up for some shortcomings of the Classical Test Theory.Parameter estimation is one of the core issues in the study of Item Response Theory,the development of Item Response Theory benefits a lot from the development of parameter estimation methods.However,in the practice of psychometrics,we often encounter missing values in the response matrices,which bring trouble to the parameter estimation.At this time,we need to take some remedial measures to minimize the impact of the missing data on the parameter estimation.In this paper,two classification methods in the field of machine learning,CART classification tree and Logistic regression,are introduced,and compared with incorrect(IN)method,personal mean imputation(PM),random imputation(RI)and full information maximum likelihood estimation(FIML).This paper explores the effects of different missing data techniques on the accuracy of parameter estimation for two-parameter logistic model under different number of examinees,different number of missing items,different missing data mechanisms and different missing rates.A comparative study of various methods is carried out by using practical data.The experimental results show that the two new methods and full information maximum likelihood estimation have high prediction accuracy and good stability.