A Comparison of χ2, RFA and IRT Based Procedures in the Detection of DIF. Gómez-benito, J. & Navas-ara, M. J. Quality and Quantity, 34(1):17–31, February, 2000.
A Comparison of χ2, RFA and IRT Based Procedures in the Detection of DIF [link]Paper  doi  abstract   bibtex   
Bias research began at the end of the 1960s anddeveloped rapidly in the following decades for obvioussocial and political reasons, and due to the importantimpact that this issue has on the field ofpsychological and educational measurement. Since then,several methods have been proposed for the study anddetection of item bias or differential itemfunctioning (DIF). This paper presents a simulationstudy comparing the potential of some of these methodsfor detecting DIF: two IRT-based techniques (area measures), three χ2-based procedures (MantelHaenszel, Logit Model and Logistic Regression) and theRestricted Factor Analysis method. The results showthat the technique that appears to do the best job isthe Mantel Haenszel statistic. Moreover, all detectiontechniques tend to overidentify DIF items, that is,some of the items labeled with DIF may in fact bewithout DIF. This tendency is slightly reversed in theLogistic Regression procedure.
@article{gomez-benito_comparison_2000,
	title = {A {Comparison} of χ2, {RFA} and {IRT} {Based} {Procedures} in the {Detection} of {DIF}},
	volume = {34},
	issn = {1573-7845},
	url = {https://doi.org/10.1023/A:1004703709442},
	doi = {10.1023/A:1004703709442},
	abstract = {Bias research began at the end of the 1960s anddeveloped rapidly in the following decades for obvioussocial and political reasons, and due to the importantimpact that this issue has on the field ofpsychological and educational measurement. Since then,several methods have been proposed for the study anddetection of item bias or differential itemfunctioning (DIF). This paper presents a simulationstudy comparing the potential of some of these methodsfor detecting DIF: two IRT-based techniques (area measures), three χ2-based procedures (MantelHaenszel, Logit Model and Logistic Regression) and theRestricted Factor Analysis method. The results showthat the technique that appears to do the best job isthe Mantel Haenszel statistic. Moreover, all detectiontechniques tend to overidentify DIF items, that is,some of the items labeled with DIF may in fact bewithout DIF. This tendency is slightly reversed in theLogistic Regression procedure.},
	language = {en},
	number = {1},
	urldate = {2020-06-02},
	journal = {Quality and Quantity},
	author = {Gómez-benito, Juana and Navas-ara, María JosÉ},
	month = feb,
	year = {2000},
	pages = {17--31}
}

Downloads: 0