ERIC Number: ED334230
Record Type: Non-Journal
Publication Date: 1991-Apr
Reference Count: N/A
"Thin" versus "Thick" Matching in the Mantel-Haenszel Procedure for Detecting DIF.
Donoghue, John R.; Allen, Nancy L.
This Monte Carlo study examined strategies for forming the matching variable for the Mantel-Haenszel (MH) differential item functioning (DIF) procedure. Data were generated using a three-parameter logistic item response theory model, with common guessing parameters. The number of subjects and test length were manipulated, as were the difficulty, discrimination, and presence of DIF in the studied item. Each replication of a test represented a test administration. Replications were made at two levels: replication within test administrations, and replication of responses to the studied items for each test administration. There were 20 replications for responses to the studied item, and 20 replications of administrations within test conditions. "Thin matching" (the use of the total score as the matching variable for MH) was compared to forms of "thick matching" (forming the matching variable by pooling total score levels). Outcome measures included the transformed log-odds delta(sub MH), standard error (delta(sub MH), and the MH chi square. The results indicate that "thick matching" can improve the performance of the MH procedure for detecting DIF. For short tests (5 to 10 items), "thin matching" yielded very poor results, with a tendency to falsely identify items as possessing DIF against the reference group. For long tests (40 items), especially with adequate sample sizes (1,600), "thin matching" yielded the best results of any method examined. Tests of intermediate lengths yielded similar results for "thin matching" and the best methods of "thick matching". Implications for the use of auxiliary information in DIF studies are discussed. Ten data tables and a 21-item list of references are included. (RLC)
Publication Type: Reports - Research; Speeches/Meeting Papers
Education Level: N/A
Authoring Institution: N/A