EXPLORATION OF CAUSES OF BIAS IN TEST ITEMS

Open Access
Publisher Website
Google Scholar
Abstract
Although the study of item bias has been an active area of research for the past few years, little progress has been made in understanding what factors in a test might create bias. To advance this understanding, a number of hypotheses were developed concerning elements of the items that might affect the performance of Blacks and Whites differently. The elements investigated were characteristics that would be common to several items that otherwise measured different concepts or content.Seven general hypotheses were developed and tested in the form of 16 specific hypotheses, most with five or six items. For purposes of this study, items were developed in pairs, with a hypothesized element present in one item and missing or modified in the other. The item pairs were assembled into two separate sections and administered as part of the Graduate Record Examinations General Test. Pairs of test sections were developed for each of the three areas tested in the General Test: verbal, quantitative, and analytical ability. These sections were spiralled for administration to yield randomly equivalent groups of examinees.The 16 specific hypotheses were evaluated separately using log linear analyses. Results showed that 10 of the 16 hypotheses showed interactions between group membership and item version indicating a differential effect of the manipulation on the performance of Black and White examinees. Eight of these hypotheses were also analyzed for item bias at the individual item level, using a procedure based on item response theory.The conclusion was that item manipulations of the type hypothesized did have a differential effect on the performance of Black and White examinees. The complexity of the effects, however, suggested that other uncontrolled factors affecting performance were also operating. A number of suggestions for future research based on these results were provided.