masoud gerami pour; mohamad reza falsafi nejhad; ali delavar; nour ali farrokhi
Volume 3, Issue 9 , October 2012, , Pages 105-122
Abstract
Although numerous methods have been proposed for detecting biased items, but a few researches have empirically investigated the power and efficiency of these methods. The main goai of this research was to apply IRT-based likelihood ratio test and confirmatory factor analysis in detecting differential ...
Read More
Although numerous methods have been proposed for detecting biased items, but a few researches have empirically investigated the power and efficiency of these methods. The main goai of this research was to apply IRT-based likelihood ratio test and confirmatory factor analysis in detecting differential item functioning (DIF) in high stakes tests. Monte Carlo simulated methods were used to answer the research questions. Required data were simulated through WINGEN2 in the form of 100 tests with 30 items that were fitted to 2PL model. Distributions of item difficulties and discrimination powers of all the tests were normal. Responses of 1000 examinees were also simulated with normal ability distribution for each test. Estimation methods of marginal maximum likelihood and weighted least squares were used to detect type and magnitudes of DIF. Data analysis in consecutive replications showed that IRT based methods were superior to CFA methods in detecting DIF. This superiority was observed in ail DIF conditions (low, moderate and high). After all. differences between two methods were small at 1000 sample sizes. No differences were observed between two methods in detecting different types of DIF. Results of this study confirm the results of Meade and Lautenschlager (2004; 2006), but is in contrast with the results of Flowers et al (2002). Finally, Likelihood Ratio test is recommended if there is any limitation in applying methods of detecting DIF.
