The Relationship Between College Freshman Class Size and Other Institutional Characteristics and the Accuracy of Freshman Grade Prediction

ACT RES EARCH REPORTNo. 8282January 1982^THE RELATIONSHIP BETWEEN COLLEGE FRESHMAN CLASS! SIZE AND OTHER INSTITUTIONAL CHARACTERISTICS AND THE ACCURACY OF FRESHMAN GRADE PREDICTIONSR. Sawyer E. J. Maxey£ THE AMERICAN COLLEGE TESTING PROGRAM THE RELATIONSHIP BETWEEN COLLEGE FRESHMAN CLASS SIZE AND OTHER INSTITUTIONAL CHARACTERISTICS AND THE ACCURACY OF FRESHMAN GRADE PREDICTIONS Prepared by the Research and Development Division The American College Testing ProgramFor additional copies write:ACT Publications P.O. Box 168 Iowa City, Iowa 52243(Check or money order must accompany request.) Price: $1.00©1982 by The American College Testing Program. All rights reserved. ABSTRACTThis report docum ents the validity of ACT test scores and self-reported high school grades in predicting college freshman grade average. The accuracy of prediction equations based on these measures is docum ented for institutions of differing freshman class size, affiliation, degree level, and racial/ethnic com position.The results in this report are based on data collected from a random sample of 205 colleges that participated in the ACT Research Services in 1974-75 and 1976-77. A separate prediction equation for each college was calculated from its 1974-75 data Each resulting prediction equation was then applied to data for the 1976-77 freshmen, and the predicted and actual grades were compared.The relationship between predictive validity and freshman class size was further investigated in two additional studies. In the first study, prediction equations were developed and cross-validated separately fo r males and females in each college. In the second study, prediction equations were developed from random subsamples of the 1974-75 freshman data from each college. Both studies supplied evidence of the relationship between prediction accuracy and sample size for samples smaller than the freshman classes represented in the data base.The predictive validity of AC T test scores and high school grades was weakly related to freshman class size at colleges with 90 or more freshmen. For example, the average mean absolute error of prediction ranged only from .51 to .54 grade units over the five size categories studied. Similarly, the average cross-validated correlation ranged from .53 to .56 over the five size categories.Prediction accuracy was moderately related to the institutional characteristics affiliation, degree level, and racial/ethnic com position. The average mean absolute error, for example, was .49 grade units for private colleges and .55 grade units for public colleges. The average mean absolute error was .55 grade units fo r two-year colleges, .50 grade units for four-year colleges, and .52 units for colleges wrth graduate programs. For colleges with the smallest proportion of black students, the average mean absolute mean error was .51 grade units, and for colleges with the highest proportion of black students, it was .59 grade units.Am ong the total group of colleges, the accuracy of separate-sex predictions was less strongly related to freshman class size than it was to the other institutional characteristics studied. The accuracy of separate-sex predictions was, however, more strongly related to freshman class size at private and four-year institutions than at other kinds of institutions.Com bined-sex equations based on simple random samples of size 50 from the base year data were almost as accurate, on the average, as equations based on all records from the colleges. These results suggest that ACT data could be used to make predictions of acceptable accuracy at colleges with as few as 50 freshmen. THE RELATIONSHIP BETWEEN COLLEGE FRESHMAN CLASS SIZE AND OTHER INSTITUTIONAL CHARACTERISTICS AND THE ACCURACY OF FRESHMAN GRADE PREDICTIONSRichard Sawyer E. James MaxeyIntroductionThe American College Testing Program (ACT) offers research services through which colleges can predict the freshman grades of future students (The American College Testing Program, 1981). The students' pre­dicted grades are based on their ACT test scores, their self-reported high school grades, and, optionally, on other predictive information. The predicted grades are calculated by weighting the predictor variables in multiple regression equations that are specific to each college.The weights in a college’s prediction equation are usually calculated from data on an entire previous freshman class (or classes). Because these weights are estimates whose accuracy depends on the size of the base sample used to calculate them, and because error in estimating