The Effect Of Anchors' Parameter Drift On Equating And The Methods To Remove It

In large-scale examination, usually the ability parameter of population participating in different test forms will be equated for comparing their measurment results on the same scale. Embedding anchors in test is frequently used in equating , which requires not only the anchors'representation in content but also equivalent function in different test forms. As a result of affecting by certain factors unrelated to the measurement objective, item parameter will change in different administration. Goldstein (1983) called this phenomenon item parameter drift (IPD).In essence, both Item Parameter Drift and Differential Item Functioning (DIF) are concerning the issue of measurement invariance. Compared with the research on DIF, there are few research on Item Parameter Drift and most of research has been mostly concerned about the decline in the difficulty parameter. At present, there is no research involved in how the equating are influeced by both difficulty and discriminating parameter drift at the same time. In the first research, comparision among mean/sigma method (M/S), the characteristics curve (TCC) and concurrent calibration (CC) are made by by simulating data, under two kind of equating deisgn, four kind of proportion of drifting anchors, and eight kind of drift mode. The research found that:(1) With the number of drifting anchors increasing, the equaing bias of three methods uually shows an upward trend. The sensitivity of three methods are different, and affected by the discrepancies in the ability of two populations.(2) In all kind of drift mode, the positive discriminating parameter drift(a+) lead to the smallest equating bias. The drift modes that the drift direction of difficulty parameter and discriminating parameter are different(a+b-, a-b+), lead to greater equating bias than the modes that the drift direction of difficulty parameter and discriminating parameter are the same (a+b+, a-b-).(3) Three equating methods'performance are affected by the drift mode. There is no a method to have absolute advantage. However, in most conditions the deletion of drifting anchors will reduce each method's equating bias.(4) In the condition of absence of drifting anchors, the equating bias of TCC is less than M/S in equivalent group design, but in non-equivalent group design and vice versa. In the two equating designs, CC lead to lowest equating bias in three methods.In the second reserch, MH test and logistic regression are used to examine the anchor in a china's large scale examination. The results found that the existence of 22 drifting anchors cause lots of them misfit in the test form of 2007. M/S method is used to compute the equating coefficients. The equating coefficients are very differnet before and after the deletion of drifting anchors. Obviously, the inclusion of drifting anchors in equating will lead to a higher level of the error and anchors'parameter drift should be examined as a regular work during equating procedure.At last ,the limitation of the paper and suggestions for further research on Item Parameter Drift are given.