There is a need to develop and investigate methods which can assess the Item Response Differences (IRD) found in all the options of an item. In this study, such an investigation was referred to as Polytomous Differential Item Functioning (PDIF). The purpose of this study was to present and investigate the performance of four new approaches in the assessment of PDIF. The four approaches are a MANOVA (MCO) and a MANCOVA (MCA) approach applied to categorical dependent variables, a Polytomous Logistic Regression (PLR) approach, and an ANOVA analysis based on the item responses quantified by Dual Scaling (DS). In this study the effectiveness of these approaches (MCA, MCO, PLR, and DS) as well as the Log-Linear (LOG) approach of Mellenbergh (1982) were assessed under various conditions of test length, sample size, item difficulty, and the amount and location of PDIF. A two-parameter polytomous logistic regression model was used to generate the data. In this study, only uniform PDIF was introduced in the alternatives of the item. The type of PDIF simulated (e.g. uniform) in this study did not allow for a direct comparison of the nonuniform test of hypothesis between the Logistic (LOG and PLR) approaches and the MAN(C)OVA (MCA and MCO) approaches because the Logistic approaches test for a difference in logits while the MAN(C)OVA approaches test for a difference in proportions. It was shown in this study that varying the probability of choosing the alternatives resulted in uniform logit differences which did not only translate into uniform differences in proportions but also translated into nonuniform differences in proportions. These differences affected the interpretation of the PDIF results because the test of nonuniform PDIF for the Logistic procedures corresponded to a valid test of the null hypothesis while the MAN(C)OVA results for nonuniform PDIF had to be adjusted in order to yield a test which approximated a true test of the null hypothesis. The results of this study lend some optimism to the employment of the MCA and PLR approaches. (Abstract shortened by UMI.)
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/9904 |
| Date | January 1994 |
| Creators | Mâsse, Louise C. |
| Contributors | Gessaroli, Marc, |
| Publisher | University of Ottawa (Canada) |
| Source Sets | Université d’Ottawa |
| Detected Language | English |
| Type | Thesis |
| Format | 211 p. |
Page generated in 0.002 seconds