NotesFAQContact Us
Collection
Advanced
Search Tips
ERIC Number: ED359214
Record Type: Non-Journal
Publication Date: 1992-Jun
Pages: 31
Abstractor: N/A
Reference Count: N/A
ISBN: N/A
ISSN: N/A
The Dimensionality of Test Data Generated by Compensatory and Non-Compensatory Two-Dimensional IRT Models and Its Effect on Model-Data-Fit.
Liu, Xiufeng
The difference between compensatory and non-compensatory item response theory (IRT) models in terms of the dimensionality of test data generated by them, and its effect on the model-data-fit were examined. The STRESS (proportion of variance not accounted for by the multidimensional scaling model) and RSQ (proportion of variance accounted for by the multidimensional scaling model) values in multidimensional scaling for unidimensional test data files were used as criteria for examining the violation of unidimensionality. The number of items significantly misfitting the unidimensional model and the mean chi-squares for all items in the unidimensional data files were calculated as reference indicators of model-data-fit of items. Compared to data generated by the compensatory model, test data generated by the non-compensatory model tend to be more two-dimensional and to more seriously misfit the three-parameter unidimensional model. Test data generated by the compensatory model tend to be over-unidimensional and to seriously misfit the three-parameter unidimensional model. Although there is a significant difference between the models in terms of dimensionality, the difference only has a significant effect on model-data-fit in terms of the number of items rejected, but has no significant effect in terms of mean chi-square values. Five tables and three figures illustrate the analyses. (Author/SLD)
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A