Tests of regression slopes or intercepts for test bias analysis.
Version: 2.1MacPotthoff performs simultaneous and separate tests of regression slopes and intercepts for test bias analysis.
Operating System: Mac OS X
Numerous methods of quantatively assessing test bias have been proposed. These include various item metrics, construct methods, content, and predictive procedures. An influential early position regarding the importance of predictive validity indicated that, "a test is considered fair for a particular use if the inference drawn from the test score is made with the smallest feasible random error and if there is no constant error in the inference as a function of membership in a particular group" (Cleary et al., 1975, p. 25).
One of the more popular methods for detecting bias of prediction was published by R. F. Potthoff (1978). Potthoff's model provides both simultaneous and separate tests of regression slopes and intercepts. If a significant F results from the simultaneous test, one may conclude that predictive bias exists and may then conduct separate tests of slopes and intercepts to ascertain whether the bias in prediction is constant (intercepts differ) or varies relative to the distance of scores from the mean (slopes differ). It is generally accepted that no test bias exists if the predictive relationship of two or more groups can be adequately described by a common regression line. If you would like to see an application of the Potthoff procedure you may find it in Weiss and Prifitera (1995).