Educators and practitioners have been striving for bias-free tests for the last few decades. As a result of this, several indices for bias detection have been developed, among which are the logistic regression and Mantel-Haenszel procedures. However, the effects of variables other than DIF on the performance of the logistic regression and Mantel-Haenszel indices have yet to be researched. The present study examines the effects of sample size, item difficulty, item discrimination, and ability distribution on the distributions and percentiles (P90 and P95) of logistic regression and Mantel-Haenszel statistics under the null hypothesis. Simulated data were used in order to evaluate the effects of the stated variables on the distributions of the logistic regression indices of uniform (LU) and nonuniform (LN) differential item functioning DIF or item bias. The same simulated data were used to evaluate the effects of the variables on Mantel-Haenzel procedure (MH-delta and MH-CHISQ). Results of this study show that the logistic regression procedure has advantages over the MH procedures, taking into account the effects of the independent variables studied. This is evident from the fact that the distribution of LN and LU index are known and that the four independent variables had no significant effect on the LU index. However the observed values were notably larger than expected values. Further research should be done to evaluate the effects of the stated variables and others such as test length, and using data with known amount of dif. Generalization of this study should be proved by replications of its findings. Evidently, variables other than DIF, significantly influence the two procedures. (Abstract shortened by UMI.)
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/10899 |
| Date | January 1992 |
| Creators | Ochieng, Charles M. O. |
| Publisher | University of Ottawa (Canada) |
| Source Sets | Université d’Ottawa |
| Detected Language | English |
| Type | Thesis |
| Format | 1992 p. |
Page generated in 0.0013 seconds