Constructing a Universal Scale of High School Course Difficulty

This study examined the usefulness of applying the Rasch rating scale model (Andrich, 1978) to high school grade data. ACT Assessment test scores (English, Mathematics, Reading, and Science) were used as " common items " to adjust for different grading standards in individual high school courses both within and across schools. This scaling approach yielded an ACT Assessment-adjusted high school grade point average (AA-HSGPA) that was comparable across schools, cohorts, and among students within the same school and cohort who take different courses. The AA-HSGPA was constructed for all ACT-tested students (N=36,652) in 50 selected high schools. First-year college grades at a large public university were available for approximately 1,500 of these students. AA-HSGPA was a better predictor of first-year college grade point average (CGPA) than the regular high school grade point average (HSGPA). As expected, the regression of CGPA on HSGPA for high schools grouped by difficulty with regard to grading policy (easy or hard) differed, but the regressions of CGPA on AA-HSGPA and the ACT Composite score (ACTC) did not. The best model for predicting CGPA included both the ACT Composite score and AA-HSGPA. As a measure of academic achievement, the grade point average (GPA) is limited by the extraneous effects of schools and courses. The problem of school effects was first to be recognized. Linn (1966) noted that variation in grading practices among high schools made the high school GPA (HSGPA) a sub-optimal predictor of achievement in college. The problem of course effects has received attention more recently. Students can earn a higher grade point average simply by taking easier courses. These problems fuel grade inflation at the high school level (Ziomek and Svec, 1995), and create incentives for students to avoid courses in difficult subjects such as mathematics and science (Johnson, 1997). In an effort to improve the prediction of college GPA (CGPA) by HSGPA, a number of earlier studies attempted to adjust both the predictor and criterion grades prediction systems, " was an attempt to control for school effects. Disappointment with this approach centered largely on the practical difficulties of its implementation. It was difficult to find enough students from a given high school within a given college to estimate coefficients in nested regression equations. Data for one or more variables in the equations were frequently missing for many of the students who could be found. Linn (1966) noted that the accuracy of central prediction …