Effects of Improving the Reliability of the GPA on Prediction Generally and on Comparative Predictions for Gender and Race Particularly

Abstract

The reliability of a method of adjusting grade point averages for differences in departmental grading standards was examined, as were the effects of such adjustments on the predictive validity of high school grades, SAT scores, and achievement test scores. The index of differential grading standards for all on-time graduates of the Dartmouth College class of 1986 was quite reliable, and its use in adjusting grade averages increased predictive validity, reduced its erosion over years, reduced the apparent underprediction of women, and improved predictions for blacks. Differential group enrollment in courses in the science division seems to account for much of the effect of adjustment on grades. Improvement in the reliability of the criterial grade averages also was shown to have similar effects on gender and race prediction in another data set,In an earlier paper (Strenta & Elliott, 1987) we showed that the prediction by Scholastic Aptitude Test (SAT) scores of first-year grade point averages (GPA1) or four-year cumulative (CUM) grade point averages were improved when the variance in the GPA criterion associated with different departmental grading standards was reduced, either by adjusting the grades by an index of differential grading rigor, or by employing as a criterion a course or courses in which grading standards were nearly uniform. The data came from all the members of the Dartmouth class of 1983 and from all psychology majors over a period of 9 years.In the present investigation, using data from all on-time graduates of the class of 1986, we extended the prior analyses in several ways. First, we speculated that adjusting for within as well as between-department standards would improve prediction, and we made such adjustments here. Second, we extended the between-department analysis from 18 to 33 departments and programs in order to make the index more complete. Third, we employed all three traditional predictors--SAT, Achievement Test scores (ACH), and high school rank (HSR)--in the analyses. Fourth, we analyzed the prediction of independently and annually computed GPAs (GPAI, GPA2, etc.) to examine the effect of criterion adjustment on the course of prediction over the undergraduate career. Fifth, we examined the effect of criterion adjustment on prediction by ethnicity and by gender, using current as well as earlier data not heretofore reported.The circumstances that occasion GPA adjustments are these: Certain depart-333

College Major and Gender Differences in the Prediction of College Grades

“…For example, in analyses of combined Latino American and non-Latino white groups, Pennock-Roman (1990, Table 3.15) found that humanities, social sciences, business, and education majors were more leniently graded 14 than physical sciences and engineering majors at all six institutions studied, but the biological/health sciences showed less consistent results. The findings from the present investigation and Pennock-Roman (1990) are consistent with many studies that have shown large contrasts in grading leniency between quantitative and nonquantitative majors (Elliott and Strenta 1988;Goldman. and Hewitt 1975;Goldman er al, 1974).…”

“…The categorization of majors used here apparently had no greater advantage in reducing gender differences as compared with the dummy-variable approach to the coding of majors into four broad categories used in the earlier analyses (Pennock-Roman 1990). The evidence from several studies (Elliott and Strenta 1988;Goldman and Hewitt 1975;Goldman et a!. 1974;Pennock-Roman 1990) suggests that the most important distinction to make is that between quantitative and nonquantitative majors.…”

“…The distribution of grades in engineering and the physical sciences tends toward higher frequencies in the range of C or below more frequently than the distribution of grades in the humanities and certain social sciences. These differences are found even when previous academic achievement and test scores are taken into account (Elliott and Strenta 1988;Goldman and Hewitt 1975;Goldman er al. 1974;Goldman and Widawski 1976;Strenta and Elliott 1987;Willingham 1985).…”

“…A number of studies have found that gender differences in the prediction of college grades (usually in the direction of underprediction of females' grades) are reduced, eliminated, or occasionally reversed when grading leniency was controlled, either by predicting individual course grades (McCormack and McLeod 1988; Rarnist et al 1994), or by adjusting the cumulative grade-point average (Elliott and Strenta 1988;Stricker et al 1991;Young 1991). Nevertheless, some gender differences in the prediction of college grades remained statistically significant even after controlling for grading standards (Stricker et al 1991), but the differences tended to be smaller or nonexistent at more selective colleges with students who had high average composites of high school grades and SAT scores (Ramist et al 1994).…”

“…Nevertheless, some gender differences in the prediction of college grades remained statistically significant even after controlling for grading standards (Stricker et al 1991), but the differences tended to be smaller or nonexistent at more selective colleges with students who had high average composites of high school grades and SAT scores (Ramist et al 1994). Some authors have argued that test bias (among other explanations) cannot be ruled out (e.g., Elliott and Strenta 1988) because gender differences have persisted in the prediction of individual course grades in psychology (Elliott and Strenta 1988), mathematics (Bridgeman andWendler 1989, 1991) and a variety of other subject areas (Ramist et al 1994).…”