Is High School GPA Better than Admission Test Scores for Predicting Academic Success in College

Abstract

This study compared the effectiveness of high school CPA and ACT score for predicting different levels of first-year college CPA. We estimated predictions of achieving first-year CPA levels ranging from 2.00 to 3.75 at 84 postsecondary institutions. Both high school CPA and ACT score were effective predictors of achieving moderate levels of first-year CPA (e.g., 2.00). High school CPA was somewhat more accurate than ACT score at predicting success at these levels. High school CPA was not an effective predictor of success at higher levels of first-year GPA, however. ACT score, in contrast, was effective in predicting success at all first-year GPA levels. College admissions officials typically use both high school GPA and scores on college entrance tests (such as the ACT Assessment and the SAT I) to predict, formally or informally, an applicant's probability of academic success in the first year of college (Breland et al. 2002). Academic success is typically measured by first-year college GPA. However, with changes in admissions policies in Texas, Florida, California, and Washington, and legal challenges to racial/ethnic preferences, many institutions are reexamining their admissions policies (e.g., Hebel 2003; Selingo 2003). Although high school GPA and college GPA both measure educational achievement, they also include other personal characteristics such as effort, attendance, conformity, and motivation (Goldman and Widawski 1976; Stiggins, Frisbie, and Griswold 1989). In contrast, scores on curriculum-based college entrance tests like the ACT Assessment primarily measure educational achievement in college-preparatory courses (ACT 1997c). One might therefore expect that both high school GPA and ACT scores are related to first-year GPA, but in different ways: High school GPA likely relates to both the cognitive and the noncognitive components of college GPA. ACT scores, on the other hand, directly relate primarily to the cognitive components of college GPA. Goldman's and others' research in the 1970S found that non-achievement components in college grades were related to student ability (e.g., Goldman and Hewitt 1975; Goldman, Schmidt, Hewitt, and Fisher 1974; Goldman and Widawski 1976).They found that high college grades were more likely to reflect cognitive achievement and were less likely to reflect noncognitive factors. One might therefore expect that: * Predictions of moderate levels of college GPA (e.g., 2.00) based on high school GPA are more accurate than predictions based on ACT scores, but that *Predictions of high levels of college GPA (e.g., 3.00) based on ACT scores are more accurate than predictions based on high school GPA. The research comparing predictions of first-year GPA from ACT scores or high school GPA has largely been correlational in nature (e.g., ACT 1998b). For example, recent summary statistics for institutions participating in ACT'S Research Services showed that across 129 colleges, the median multiple correlation (across institutions) between first-year GPA and the four ACT scores (in English, mathematics, reading, and science) was 0.43 (Maxey, forthcoming). The correspondingmedian correlation between first-year GPA and four high school average grades was 0.48.The eight-predictor ACT score/ high school grade average multiple correlation with first-year GPA was 0.53. Correlations, however, pertain to the entire range of college GPA, and not to particular levels of GPA. The purpose of this study is to compare the accuracy of high school GPA and ACT Composite scores for predicting achievement of successively higher levels of first-year college GPA. The levels considered were first-year GPAS of 2.00, 2.50, 3.00, 3.25, 3.50, and 3.75 or higher. In practice, prediction equations based on data from one freshman class are applied to the test scores and/or high school GPAS of future applicants. Because students may differ over time in their test scores, high school GPAS, or college grades, predictive validity statistics developed from one year's data may misstate the strength of the relationship associated with actual use of such predictions. …

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 74 citations exist upstream.