Clauser J. — American Board of Internal Medicine
Presented: National Council on Measurement in Education Annual Meeting, April 2015
In a nonequivalent groups anchor test (NEAT) design, score scales are linked across the administration of two examinations by fixing item parameters for a set of common items known as an anchor. In principle, the most accurate linking constants are achieved when all common items are used in the anchor. In practice, however, when item parameters drift over time, omitting these items may provide a more accurate link. Many methods exist for assessing the magnitude of this item parameter drift. Unfortunately, these methods do not provide an objective criterion for excluding items from the anchor. Instead, practitioners are typically forced to rely on professional judgment or relativistic screening rules when identifying item drift. This paper uses estimated error in examinee ability as a criterion for evaluating anchors. A method for optimizing this criterion is presented.
For more information about this presentation, please contact Research@abim.org.