INTERACTIONS BETWEEN ITEM CONTENT AND GROUP MEMBERSHIP ON ACHIEVEMENT TEST ITEMS

The three-parameter logistic model has been used to identify items that may be considered biased. Application of this technique is limited by the requirement of large samples for obtaining estimates, often much larger than is feasible for some subgroups that are of particular interest. An approach is presented for using the three-parameter logistic model that is applicable even when one of the subgroups is relatively small. The approach is illustrated with a mathematics test. In the discussion of the results it is suggested that the interactions between item content and group membership, commonly called item bias, may generally be conceptualized as indications of multidimensionality. It is also suggested that "item bias" for an achievement test may sometimes more appropriately be called "instruction bias."