Group Differences in Structured Tests

Abstract

During the last two decades much effort has been put into developing and refining procedures to detect bias in test items. Bias is defined here as any effect on an observed score which is not found to an equal extent in the psychological universe to which the scores are generalised (cf. Poortinga and Van de Vijver, in press). The vast majority of publications on this subject has appeared in methodologically oriented journals; remarkably few references are made to the bias tradition in journals implicitly devoted to intergroup comparisons. Apparently, the study of item bias has neither created nor met a strong demand among cross-cultural psychologists. Although it can be argued that this is simply a matter of time, it seems more likely that the item bias tradition will influence applied research only to a limited degree because of the conceptual rigidity of the approach. Item bias procedures are post hoc procedures meant to detect items with differential statistics across groups. When a test is administered to people of two cultures and differences in endorsement rates are the same for all items, except for one item which shows a larger difference, this latter item is typically considered to be biased. Items are dichotomously labelled as biased or unbiased.