This study investigated how the collapsing of categories impacted the information function and the convergence of the iterative item calibration in polychotomous Item Response Theory. The scores of 1000 examinees on twenty-four twelve-item performance assessment test batteries were simulated. The experimental factors were direction of category collapsing (upward and downward), three levels of item difficulty ($-$1.0, 0.0, and 1.0), three levels of item discrimination (0.4, 0.9, and 1.6), and two levels of inter-rater reliability (.90 and.95). PARSCALE was used to calibrate the tests and provide the information data. Factorial repeated measures analysis was used on the three experimental designs for maximum item information, total item information, and EM-cycles. / The results demonstrated that (1) combining raters' evaluations reduced the effective item discrimination of an item and increased the range of step difficulties, but left item difficulty essentially unchanged; (2) overall, the collapsing of categories increased item discrimination, reduced the number of EM-cycles and total information, and left item difficulty and maximum information essentially unchanged; and (3) within the limited range of inter-rater reliability studied, the "high" level of inter-rater reliability did not provide more information than the "low" level. / All three designs were interactive, with significant two-way interactions in the information designs and a three-way interaction in the EM-cycles design. Within the experimental criteria of statistical significance and practical importance, it was demonstrated that the "high" levels of the experimental factors contributed to the few significant reductions in information and the number of EM-cycles that were detected. The direction of collapsing only had a significant effect in the EM-cycles design. / Because performance assessment batteries typically are composed of items which span the range of the item characteristics used as experimental factors in this study, it was concluded that the overall effects are more germane to practical applications than the few significant effects. Based on these results, practitioners who find it useful to collapse categories under the conditions considered here may do so without any expected adverse effects on maximum and total information. / Source: Dissertation Abstracts International, Volume: 55-03, Section: A, page: 0541. / Director: Richard Tate. / Thesis (Ph.D.)--The Florida State University, 1994.
| Identifer | oai:union.ndltd.org:fsu.edu/oai:fsu.digital.flvc.org:fsu_77112 |
| Contributors | Lecointe, Darius Alexander., Florida State University |
| Source Sets | Florida State University |
| Language | English |
| Detected Language | English |
| Type | Text |
| Format | 210 p. |
| Rights | On campus use only. |
| Relation | Dissertation Abstracts International |
Page generated in 0.0016 seconds