Possible bias in selection procedures used for employment and college admissions is of crucial social and educational importance. However, there are many different definitions of what constitutes bias with each definition based on different values and with different implications for how selection should be accomplished. A number of these definitions of bias and their implications are examined, and a new conditional probability model of fairness based on equal opportunity for potentially successful applicants is presented. This conditional probability model is proposed as an intuitively appealing and socially desirable model for use in many selection situations in employment and college admissions.

As strongly suggested by recent work, patterns of gender difference can change because of changes in the selectivity of the sample itself. This is a statistical influence connected with the distributions of female and male scores, rather than a substantive influence related to demographic characteristics of the sample such as age or ethnicity. It is, nonetheless, an important influence because gender differences that are partly statistical in origin can easily confuse possible implications regarding education and assessment.This report is part of a larger project directed by Warren Willingham and Nancy Cole on gender differences and similarities in achievement and implications for fair assessment. We are grateful to Nancy Cole for useful conversations about the complex topic in which the particular problem here addressed is imbedded. We wish to thank a number of reviewers: the project advisory committee (and also, Larry Hedges for his careful review. We are especially grateful to several colleagues for their assistance: to Judy Pollack for help with the analyses, to Linda Johnson for several contributions to the project and to Joan Stoeckel and Carol Crowley for help with the manuscript.

No abstract

For most of the 20th century, measurement professionals paid little interest to item and test fairness. A confluence of events in the late 1960s and early 1970s led to an intense interest in fairness issues among measurement professionals. In spite of more than 30 years of effort, there is still no generally accepted definition of fairness with respect to testing and no measure that can prove or disprove the fairness of a test. To advance the fairness of tests, measurement professionals must pay more attention to reducing group differences at the design stage of test development, to providing all examinees an opportunity to demonstrate their knowledge and skills, to deterring test misuse, and to accommodating differences among individuals.

The problem of test bias has recently received tremendous scientific and public scrutiny. Cole reviews the approaches that have been undertaken to detect cultural, content, predictive, and selection bias in mental tests. This includes analysis of subtle differences in the content of test items to which individuals react differently and the implications of statistical differences in predictions from test scores. She argues that questions of bias are fundamentally questions of validity. A distinction is made between validity on one hand, and the question of whether a test should be used, even if valid, on the other. The author concludes that although on the technical side many things have been learned about the details of test bias, such research has not provided answers to social policy questions that must be decided regardless of whether tests are involved. -The Editors Some of the most prominent issues associated with testing in recent years have involved questions of test bias. The possibility of bias in tests has been a major focus of test critics, the courts, test developers, and scholars of testing alike. The focus has resulted in much public debate and scholarly writing, but a large gap continues to exist between the public concerns and the concerns of technical scholars of testing.The issue of test bias has gained importance as much because of the social policy issues with which it has become intertwined as because of its own intrinsic importance. Shepard (Note 1) aptly noted one prominent issue associated with test bias:One reason that bias in mental testing is so volatile an issue is that it involves the specter of biological determinism, i.e., whether there is a large difference in intelligence (IQ) between black and white Americans which can be attributed largely to inherited differences, (p. 5)The reasons for observed group differences in test scores and the possible social policy implications of different reasons are but one set of the volatile issues with which the issue of test bias has become at least marginally related. Others include issues such as the provision of educational and job opportunities for minorities (Novick, this issue; Tenopyr, this issue), the appropriateness of particular

An improved Bayesian method, due to Lindley, for the simultaneous estimation of multiple regressions in m groups is studied by applying the method to test data. Evidence is found to support the belief that in many testing applications the collateral information obtained from each subset of m − 1 colleges will be useful for estimating the regression in the mth college, especially when the sample sizes are small. Using a 25 per cent sample, the Bayesian prediction equations achieved an average 9.7 per cent reduction in mean square error, as compared with the within‐group least squares equations, when cross‐validated with a later sample. More importantly, the mean square error for the Bayesian equations based on the 25 per cent sample was only barely greater than that for the least squares equations based on the full sample data. Thus the main virtue of the method is that it permits predictions to be made separately for relevant subpopulations (e.g. male‐female) where sample sizes would otherwise be too small to achieve an acceptable degree of accuracy.

Compared the internal structural relationships of scales from the svib, the kuder occupational interest survey, holland's vocational preference inventory, the minnesota vocational interest inventory, and the american college testing program vocational interest profile. The configurations of the scales for all the inventories were found to be similar and to conform to the circular configurations of interest proposed by A. Roe and J. Holland. The common configuration of vocational interests was used to reconcile previous contradictory research results about the comparability of interest scores from various instruments and as a basis for counselor interpretation. (33 ref.)

No abstract