The performance of three popular procedures for detecting differential item functioning (DIF), the Mantel-Haenszel (MH), the Simultaneous Item Bias (SIB), and the Logistic Regression (LR) procedures were investigated and compared in three different studies. The first study compares the MH and the SIB procedures with respect to their Type I error rates and power to detect uniform DIF. Data for the study were simulated to reflect a variety of conditions. The results revealed that both the MH and the SIB procedures were equally powerful in detecting uniform DIF under most of the studied conditions. The SIB procedure showed higher detection rates than the MH procedure as the ability distribution differences increased. The second study investigated the distributions of the SIB and two variations (with and without the continuity correction in the MH statistic), to determine whether or not their distributional assumptions held. The results showed that the SIB statistic generally had the expected distributions when the sample size of the reference and the focal groups exceeded 200. The distributions assumptions of the MH statistic without the continuity correction were more readily met than those of the MH statistic with the continuity correction for all the studied conditions. The third study investigated the MH, the SIB, and the LR procedures with respect to their Type I error rates and power to detect non-uniform DIF. Data for the study were simulated under a variety of conditions. The results revealed that both the SIB and LR procedures were equally powerful in detecting non-uniform DIF under most conditions. The MH procedure was not very effective in identifying non-uniform DIF items that showed disordinal interactions. The investigation of the Type I error rates in all the three studies showed that they were within the expected limits for the MH procedure, higher than expected for the SIB and LR procedures with the SIB results showing an overall increase of about 1% over the LR results. With respect to power, the results show that the MH statistic was very effective in detecting only uniform DIF; both the SIB and LR procedures were very effective in detecting uniform as well as non-uniform DIF.
| Identifer | oai:union.ndltd.org:UMASS/oai:scholarworks.umass.edu:dissertations-9054 |
| Date | 01 January 1995 |
| Creators | Narayanan, Pankaja |
| Publisher | ScholarWorks@UMass Amherst |
| Source Sets | University of Massachusetts, Amherst |
| Language | English |
| Detected Language | English |
| Type | text |
| Source | Doctoral Dissertations Available from Proquest |
Page generated in 0.015 seconds