Extent of Overlap among Four Item Bias Methods

Abstract

Four methods of item bias detection-transformed item difficulty, item discrimination expressed as Clemans' lambda, ChiSquare, and the three-parameter item characteristic curve-were studied to determine the degree of correspondence among them in identifying biased and unbiased items in reading and math subtests of the 1978 SRA Achievement Series. Intercorrelations among the four methods were moderate at best, confirming previous research involving different item bias analysis techniques. The item discrimination method showed the least correspondence with the other three methods. The extent of overlap between the four item bias methods in identifying biased items depends on the extent of bias in the items comprising the initial pool of items. That is, except for the item discrimination method, the item bias procedures identify similar sets of most biased items.