Test Bias and Construct Validity / Arthur R. Jensen.

The several statistical methods described for detecting test bias in terms of various internal features of a person's test performances and the test's construct validity can be applied to any groups in the population. But the evidence regarding groups other than U.S. blacks and whites is either lacking or is still too sketchy to permit any strong conclusions. The evidence regarding black-white comparisons, however, is based on a number of well-known, widely used, and quite diverse standardized individual and group tests of intelligence given to a large representative sample of whites and blacks. The results are unequivocal: none of the several subjective indices of cultural bias shows any significant indication of bias in any of these tests when they are used with blacks and whites. Correlation of raw scores with age, internal consistency reliability, rank order of item difficulty, relative difficulty of adjacent items, item correlation with total score, loadings of items or tests on the general factor, and relative frequencies in choice of error distractors--all are substantially the same in black and white groups. It is concluded that these standardized tests of intelligence--the Peabody Picture Vocabulary, Raven's Progressive Matrices, Wechsler Intelligence Scale for Children, Stanford-Binet, Wonderlic Personnel Test, and most likely other similar tests--are not at all culturally biased for blacks and whites. They behave statistically the same in both racial groups and do essentially the same job in both groups. (Author/DEP)

Microfiche. [Washington D.C.]: ERIC Clearinghouse microfiches : positive.