No abstract

It is common practice to assume that the dependent variables in differential prediction studies, and analysis of variance and co-variance designs are characterized by certain psychometric properties which are invariant across the subgroups of interest. More specifically, the accuracy of the results of such analyses depends on the assumption that the dependent variables are measuring the constructs in the same metrics with equivalent reliabilities across all subgroups. Systematic procedures are outlined for testing these assumptions.

The definition of test bias-the inventory of the ways in which the term is used-has many widely disparate aspects frequently stemming from entirely different universes of discourse. This article attempts a review of the status of each of these. It seems essential to keep all of these various aspects in mind, for we continually run the risk of losing perspective on our research when we settle on one operational definition of test bias and then proceed to forget that it is only that.No matter what definition we use, because the concept is a public one we are never going to encompass all that it contains. The research on these various aspects is some of the more exciting and significant being done in psychology today, but let us not be confused about whether one of them is the real issue. As it happens, they are all the real issue.

Among the four participating ethnic groups in Project Access for Los Angeles (Black, White, Mexican-American, and Oriental), the patterns of test performance that emerged were surprisingly similar to those from two previous studies of first-grade children, even though Access involved eleventh graders. Partially replicated was the finding that overall level of performance varied with social class, while the patterns remained rather constant within ethnic group, regardless of social class.This new evidence adds to the strength of the argument that our educational systems should be attending to, and utilizing, the variety of talents and aptitudes presented by the students, rather than attempting to treat everyone as a member of the majority group.

SUMMARY In this study, a substantial number of items that were appreciably more difficult for Black than for White students were eliminated from a verbal test and from a mathematical test. Shortening the tests by this method had relatively little effect on score differences between Black and White students. This method of selecting items for elimination resulted in tests that were decidedly more difficult for both student groups. These outcomes clearly limit the value of this method for use in developing tests.