The common approach to studies of predictive bias is analyzed within the context of a conceptual model in which predictors and criterion measures are viewed as fallible indicators of idealized qualifications. (Author/PN)

Presents the drastic effects that selection on several variables can sometimes have on the validity of predictor variables. Discusses practical implications of correcting correlations for selection effects and reporting results of validity studies. (Author)

The internal validity of the RMC models, especially Model A, is examined. Concern centers on limiting evaluation to cognitive outcomes, using constant percentile as the no-treatment expectation, and using norms for one test to establish the expected no-treatment performance level for another test. (MH)

The three RMC models endorsed by the U.S. Office of Education for the evaluation of Elementary and Secondary Education Act Title I programs are based on narrowly conceived approaches to evaluation--the measurement of cognitive achievement gains. Each model requires the comparison of observed student performance with an estimate of what level of…