Çoklu atama yöntemlerinin Rasch modelleri için performansının benzetim çalışması ile incelenmesi
No Thumbnail Available
Files
Date
2012
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Sağlık Bilimleri Enstitüsü
Abstract
The aim of this study was to investigate the effects of missing data and imputation via multiple imputation methods on person and item parameter estimates obtained from Rasch models. Among from the multiple imputation methods those are available for scale data were selected and investigated for partial credit model which is a frequently used Rasch model. In addition to the multiple imputation methods, the performance of Rasch models using only available data at hand was evaluated.Simulation study was carried out and data were simulated based on partial credit model. Missing data were generated as completely at random, at random and not at random with proportions 0.10, 0.30 and 0.50. Missing data were imputed with multiple imputation based on response function, corrected item mean substitution, multivariate normal distribution and chained equations. Response function found to have minimum bias and mean square error among imputation methods. Maximum likelihood estimations from partial credit model in the missing data case were close to those when there are no missing data. The effects of both missing data and imputing data via multiple imputation methods on differential item functioning and local item independence were trivial. While standard errors of parameter estimates in the missing data case differed from those were estimated using full data, after multiple imputation, standard errors from completed data were close to those from full data. Real data application gave similar results to those found from simulation study.Consequently, although maximum likelihood estimations from Rasch models in the missing data case were unbiased, standard errrors of parameters estimates were different from those of full data. Response function imputation had the highest performance among the multiple imputation methods and for Likert type scales it can be used when the proportion of missing data is less than 0.50 and when there is a concern about precision of the estimates.
Description
Keywords
Çoklu atama yöntemleri, Rasch modelleri