ERIC Number: ED331876
Record Type: RIE
Publication Date: 1991-Apr
Reference Count: N/A
Examination of Various Influences on the Mantel-Haenszel Statistic.
Clauser, Brian E.; And Others
Item bias has been a major concern for test developers during recent years. The Mantel-Haenszel statistic has been among the preferred methods for identifying biased items. The statistic's performance in identifying uniform bias in simulated data modeled by producing various levels of difference in the (item difficulty) b-parameter for reference and focal groups was studied. Data were simulated for a 75-item test (59 non-biased and 16 biased items) with five administrations and 1,000 examinees in each group. Levels of difference in difficulty were crossed with four levels of the (item discrimination) a-parameter, five levels of the b-parameter, and two ability distributions. The results show significant effects for the difference in difficulty levels, and for the a-parameter and b-parameter values. Data suggest that: the Mantel-Haenszel statistic may differentially favor identification of more highly discriminating items, and the procedure is likely to be unable to detect bias in very difficult items. Recommendations for practitioners are provided. Four tables and four graphs present the simulated data. (Author/SLD)
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Authoring Institution: N/A
Identifiers: Mantel Haenszel Procedure
Note: Paper presented at the Annual Meeting of the American Educational Research Association (Chicago, IL, April 3-7, 1991).