Investigating the Stability of Four Methods for Estimating Item Bias [microform] / Carole L. Perlman and Others.
The reliability of item bias estimates was studied for four methods: (1) the transformed delta method; (2) Shepard's modified delta method; (3) Rasch's one-parameter residual analysis; and (4) the Mantel-Haenszel procedure. Bias statistics were computed for each sample using all methods. D...
Saved in:
| Online Access: |
Request ERIC Document |
|---|---|
| Main Author: | |
| Format: | Microfilm Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
Distributed by ERIC Clearinghouse,
1988.
|
| Subjects: |
| Summary: | The reliability of item bias estimates was studied for four methods: (1) the transformed delta method; (2) Shepard's modified delta method; (3) Rasch's one-parameter residual analysis; and (4) the Mantel-Haenszel procedure. Bias statistics were computed for each sample using all methods. Data were from administration of multiple-choice items from the Chicago Minimum Proficiency Skills Item Bank to 54,896 eighth graders, including Blacks, Whites, and Hispanics. The item bank consists of about 1,000 minimum competency items from which an annual form of 63 items is assembled. This study was based on the 46 items that were common to the 1984 and 1985 forms of the test. Thirty samples of each of 600 to 2,000 subjects were randomly drawn from the study population and tested over a 2-year period. Emphasis was whether bias related to socioeconomic differences existed. When individual group sizes were 666 or larger, the reliability of item bias indicators was quite adequate. For smaller groups, it was sometimes a problem. No one method consistently flagged more or fewer items across comparisons or the sample sizes. There was general agreement among methods on the proportion of times each item was indicated as being biased. Six tables provide the study data. (SLD) |
|---|---|
| Item Description: | ERIC Note: Paper presented at the Annual Meeting of the National Council for Measurement in Education (New Orleans, LA, April 5-9, 1988). ERIC Document Number: ED296003. |
| Physical Description: | 21 pages. |
| Reproduction Note: | Microfiche. |
| Action Note: | committed to retain 20240101 20490101 Alliance Shared Trust https://www.coalliance.org/shared-print-archiving-policies |