An Experimental, Exploratory Study of Causes of Bias in Test Items

This study evaluated 16 hypotheses, subsumed under 7 more general hypotheses, concerning possible sources of bias in test items for black and white examinees on the Graduate Record Examination General Test (GRE). Items were developed in pairs that were varied according to a particular hypothesis, with each item from a pair administered in different forms of an experimental portion of the GRE. Data were analyzed using log linear methods. Ten of the 16 hypotheses showed interactions between group membership and the item version indicating a differential effect of the item manipulation on the performance of black and white examinees. The complexity of some of the interactions found, however, suggested that uncontrolled factors were also differentially affecting performance.