NotesFAQContact Us
Search Tips
ERIC Number: ED371019
Record Type: Non-Journal
Publication Date: 1994-Apr
Pages: 21
Abstractor: N/A
Reference Count: N/A
Robustness of Unidimensional IRT Calibration in the Presence of Essential Dimensionality.
Wang, Yu-Chung Lawrence
The first purpose of this study was to investigate the stability of two essential dimensionality measures across 10 random samples within a particular assessment item (AT1) selection. Other purposes were to investigate the discrepancy of the essential unidimensionality estimates for a test across different AT1 selections and sample sizes and to investigate the validity of replacing the item response theory (IRT) unidimensionality assumption with the essential unidimensionality assumption using the existence of the invariance property of item parameters as a criterion. Results indicate that the stability of two essential unidimensionality measures is low for some tests across 10 random samples, but the correlation is high within the same sample. The essential dimensionality results for four tests across four different AT1 assignments were also different, indicating that the essential dimensionality estimate for a test is related to the characteristics of AT1 items. It was found that reducing sample size or reducing the number of test items and AT1 items does not assure unidimensionality. Relationships between the existence of the item invariance property and the essentially unidimensional item calibrations are low across test forms and mathematics areas. A further study of the criteria of AT1 items is needed to enhance the validity of replacing the IRT unidimensionality assumption by the essential unidimensionality assumption. Fourteen tables present the data. (Contains 13 references.) (Author/SLD)
Publication Type: Reports - Research; Speeches/Meeting Papers
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A