Who Will Succeed in College? When The SAT Predicts Black Students' Performance

African American students feel greater than usual concern over whether the Student Aptitude Test (SAT) gives a true reading of academic aptitude but have little understanding of when the SAT does and does not predict the grades they will achieve in college. Clearer insights about predictive validity issues for them may come from several nontraditional studies based on academic and psychosocial data. While educational underachievement prior to test-taking (Bracey, 1993; Horn & Carroll, 1997; Smith & Choy, 1995a, 1995b; Young & Smith, 1997) and issues activated during test taking are important (Jencks & Phillips, 1998; Steele & Aronson, 1995; Watson, 1972), this discussion focuses on the consequences of a given test score for future success in college. Data from several sources provide analyses of predictive validity by dominant race of the college and by gender. Analyzing SAT correlates sheds new light on the problem of predictive validity in minority populations.

Predictive Validity for White and Black Students

Within certain important limits, the SAT tells us which White students will succeed in college, despite charges that test scores do not reflect ability and do not help colleges make better selection decisions (Crouse & Trusheim, 1988; Neill & Medina, 1989). Nonetheless, the SAT-College GPA correlations among majority students are consistent. While prediction from SAT scores to grades usually includes high school grades and may lead to complex prediction equations, the SAT-College GPA correlation is the essential predictive validity statistic. Fleming and Garcia's (1998) review of 12 studies of predictive validity among White students, which included studies conducted for up to thirteen years with up to eleven different samples, found that the average correlation was 0.342. The square of the correlation indicates the amount of variance accounted for by the SAT in college GPA; in this case the variance accounted for was 11.7%. (See Fig. 1.) With the exception of a study by Pennock-Roman (1990) which produced generally low correlations, the remaining eleven studies were distinguished by the consistency with which test scores showed either moderate or strong ability to predict college grades. In short, few measures rival test scores in consistency of prediction. Virtually the only other measure able to indicate how a student will fare in college is the high school grade point average.

SAT scores may not give the same reading of future success for African American students because the predictive validity is lower than for White students. However, the evidence for lower Black predictive validity is actually inconsistent and does not fall neatly into a single category. Some researchers have found strong positive SAT-GPA correlations for Black students (e.g., Breland, 1978; Morgan, 1990), while other authors have reported that the SAT score bore no relationship to grades (e.g., Boyd, 1977; Miller & O'Connor, 1969), that students performed better than their test scores would indicate (Houston, 1983), and that students performed worse than their test scores would indicate (Breland, 1978; Crouse & Trusheim, 1988; Nettles, Thoeny, & Gosman, 1986; Kane, 1998; Temp, 1971; Vars & Bowen, 1998). Indeed, contrary to popular opinion, over-prediction (performing worse) is the most consistent occurrence in Black predictive validity studies, and few cogent explanations have been offered.

Fleming and Garcia (1998) examined eight validity studies reporting correlations for Black students (Fig. 1). The average correlation was 0.315, accounting for an average of 9.9% of the variance in college grades. While this figure is lower than that for majority students, it is a difference of only 1.8%--not as low as opinion would suggest. This difference does not support an argument for differential predictive validity. Jencks (1998) comes to a similar conclusion: that test scores have a moderate correlation with grades and that test scores predict a little better for Whites than for Blacks. Fleming and Garcia (1998) also found that the variability in correlations was substantially greater for Black student samples, where correlations ranged from...